Method for promoting the creation and use of visual content

Abstract

A system, method, and non-transitory computer-readable medium are provided for attributing generated visual content to training cases. A first visual content generated using a generative model is received. The generative model is associated with a plurality of training cases. Each training case is associated with visual content. Characteristics of the first visual content are determined. Each visual content associated with a training case is analyzed to determine characteristics of the visual content. Characteristics of the first visual content and characteristics of visual content associated with a plurality of training cases are used to attribut the first visual content to a subgroup of the plurality of training cases. The visual content associated with the subgroup of the plurality of training cases is associated with a resource. A data record associated with the resource is updated based on the attribution.

Description

[Technical field]

[0001] This application claims priority to U.S. Provisional Patent Application No. 63 / 279,111, filed November 14, 2021, the entirety of which is incorporated herein by reference.

[0002] This disclosure relates generally to the field of generative models. More particularly, this disclosure relates to systems and methods for generating visual content using generative models, and uses of the generated visual content. [Background technology]

[0003] A huge amount of visual content, such as images and videos, is created, edited, and viewed by different entities. Currently, much of the visual content is modified to accommodate different standards. However, it still requires significant efforts to achieve high-quality results when modifying visual content. Moreover, it is still difficult to select optimal modifications for different purposes. This is especially true when many items of visual content require modification. Summary of the Invention [Problem to be solved by the invention]

[0004] Therefore, there is a need to automate the modification of visual content. [Means for solving the problem]

[0005] SUMMARY OF THE DISCLOSURE Embodiments of the present disclosure provide systems, methods, and non-transitory computer-readable media for visual content generation.

[0006] In some examples, a system, method, and non-transitory computer-readable medium are provided for attributing generated visual content to training cases. For example, a first visual content generated utilizing a generative model can be received. The generative model can be a result of training a machine learning algorithm utilizing a plurality of training cases. Each training case of the plurality of training cases is associated with visual content. One or more characteristics of the first visual content are determined. For each training case of the plurality of training cases, the visual content associated with the training case can be analyzed to determine one or more characteristics of the visual content associated with the training case. The one or more characteristics of the first visual content and characteristics of the visual content associated with the plurality of training cases can be utilized to attribute the first visual content to at least one, but not all, first subgroup of the plurality of training cases. Further, it can be determined whether at least one visual content associated with a training case of the first subgroup is associated with at least one first resource. In one example, for each resource of the at least one first resource, a data record associated with the resource is updated based on the attribution.

[0007] In some examples, a system, method, and non-transitory computer-readable medium are provided for identifying visual content to be utilized in training an inference model. For example, distinctive visual content is received. Additionally, data based on at least one parameter of the inference model is received. The inference model can be a result of training a machine learning algorithm using a plurality of training cases. Each training case of the plurality of training cases can include visual content. The data and distinctive visual content can be analyzed to determine a likelihood that the distinctive visual content is included in at least one training case of the plurality of training cases. A digital signal indicative of a likelihood that the distinctive visual content is included in at least one training case of the plurality of training cases is generated.

[0008] In some examples, a system, method, and non-transitory computer-readable medium are provided for converting a non-photorealistic virtual environment into a photorealistic virtual environment. For example, a first digital signal is received that represents virtual content within an augmented reality environment. The first digital signal can be utilized to identify non-photorealistic portions of the virtual content. Further, a generative model can be utilized to analyze the first digital signal and generate a photorealistic version of the identified non-photorealistic portions of the virtual content. In one example, a second digital signal is generated that is configured to cause a wearable augmented reality device to present the generated photorealistic version in place of the identified non-photorealistic portions of the virtual content within the augmented reality environment.

[0009] In some examples, a system, method, and non-transitory computer-readable medium are provided for determining realism of artificially generated visual content. For example, one or more artificially generated visual content can be accessed. The one or more artificially generated visual content can include a particular artificially generated visual content. Furthermore, one or more captured visual content can be accessed. Each visual content of the one or more captured visual content can be visual content captured using an image sensor from a real physical environment. For each person of the plurality of people, a mix of the plurality of visual contents can be presented to the person, the mix of the plurality of visual contents can include at least one visual content of the one or more artificially generated visual content and at least one visual content of the one or more captured visual content, and the mix of the plurality of visual contents can include the particular artificially generated visual content. For each person of the plurality of people, a response to the presentation can be received from the person, and for each visual content of the group of at least one visual content mix, the response can indicate whether the person believed the visual content was artificially generated visual content. Based on the response, the realism of the particular artificially generated visual content is determined.

[0010] Consistent with other disclosed embodiments, a non-transitory computer-readable medium can store a software program including data and / or computer-implementable instructions that, when executed by at least one processing device, cause the at least one processing device to perform one or more of the methods and / or steps described herein.

[0011] The foregoing general description and the following detailed description are exemplary and explanatory only and are not intended to be limiting of the scope of the claims.

[0012] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate several embodiments of the present disclosure and, together with the following description, serve to explain the principles of these embodiments disclosed herein. [Brief description of the drawings]

[0013] [Figure 1A] FIG. 1 illustrates an exemplary communication system. [Figure 1B] FIG. 1 illustrates an exemplary communication system. [Figure 2A] FIG. 1 is a simplified block diagram of one possible implementation of a computing device. [Figure 2B] FIG. 1 is a simplified block diagram of one possible implementation of a computing device. [Figure 3A] FIG. 2 illustrates exemplary visual content. [Figure 3B] FIG. 2 illustrates exemplary visual content. [Figure 4A] FIG. 1 is a flow diagram of an example method for attributing generated visual content to training examples. [Figure 4B] FIG. 1 is a flow diagram of an example method for attributing generated visual content to training examples and selecting visual content for a particular application based on attribution. [Figure 4C] FIG. 1 is a flow diagram of an example method for attributing generated visual content to training examples and selecting visual content for a particular application based on attribution. [Figure 4D] FIG. 1 is a flow diagram of an example method for attributing generated visual content to training examples. [Figure 4E] FIG. 1 is a flow diagram of an example method for attributing generated visual content to training examples. [Figure 5A] FIG. 1 is a flow diagram of an example method for identifying visual content utilized in training an inference model. [Figure 5B] FIG. 1 is a flow diagram of an example method for identifying visual content utilized in training an inference model. [Figure 5C] FIG. 1 is a flow diagram of an example method for identifying visual content utilized in training an inference model. [Figure 5D] FIG. 1 is a flow diagram of an example method for identifying visual content utilized in training an inference model. [Figure 6A] FIG. 2 illustrates exemplary visual content. [Figure 6B] FIG. 2 illustrates exemplary visual content. [Figure 6C] FIG. 2 illustrates exemplary visual content. [Figure 6D] FIG. 2 illustrates exemplary visual content. [Figure 6E] FIG. 1 illustrates an exemplary augmented reality environment. [Figure 7] FIG. 1 is a flow diagram of an exemplary method for converting a non-photorealistic virtual environment into a photorealistic virtual environment. [Figure 8] FIG. 2 illustrates an exemplary user interface. [Figure 9A] 1 is a flow diagram of an exemplary method for determining realism of artificially generated visual content. [Figure 9B] 1 is a flow diagram of an exemplary method for determining realism of artificially generated visual content. [Figure 9C] 1 is a flow diagram of an exemplary method for determining realism of artificially generated visual content. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] Unless otherwise specified, and as will become apparent from the following description, the terms "processing," "calculating," "operating," "determining," "generating," "setting," "configuring," "selecting," "defining," "applying," "obtaining," "monitoring," "providing," "identifying," "dividing," "classifying," "analyzing," "associating," "extracting," "storing," "receiving," "transmitting," and the like, as used throughout this specification, include computer processes and / or actions that transform and / or manipulate data into other data, where the data represent physical quantities, e.g., quantities of electrons, and / or where the data represent physical objects. Terms such as "computer," "processor," "controller," "processing unit," "computing device," and "processing module" are intended to be expansively interpreted to include any electronic device, component, or unit having data processing capabilities, including, by way of non-limiting example, a personal computer, a wearable computer, a tablet, a smartphone, a server, a computing system, a cloud computing platform, a communications device, a processor (e.g., a digital signal processor (DSP), an image signal processor (ISR), a microcontroller, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a central processing unit (CPU), a graphics processing unit (GPU), a visual processing unit (VPU), etc.), possibly including embedded memory, a single core processor, a multi-core processor, a core within a processor, any other electronic computing device, or a combination thereof.

[0015] Operations according to the teachings described herein are performed by a computer specially constructed or programmed to perform the functions described.

[0016] As used herein, phrases such as "for example," "such as," "one example," "another example," "different examples," "several examples," "additional examples," "still other examples," and variations thereof, describe non-limiting embodiments of the subject matter disclosed herein. "In one instance," "several instances," "other instances," and variations thereof described herein mean that a particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment of the subject matter disclosed herein. Thus, appearances of phrases such as "in one instance," "several instances," "other instances," and variations thereof do not necessarily refer to the same embodiment. As used herein, the term "and / or" includes any and all combinations of one or more of the items listed in the associated list.

[0017] Certain features of the subject matter disclosed herein, which are, for clarity, described in the context of separate embodiments, may also be provided in combination in a single embodiment. Conversely, various features of the subject matter disclosed herein, which are, for clarity, described in the context of a single embodiment, may also be provided in the context of separate embodiments or in any suitable sub-combination.

[0018] The term "image sensor" is recognized by those skilled in the art and refers to any device configured to capture an image, a series of images, video, etc. This includes sensors that convert optical input into an image, which can be visible light (as in a camera), radio waves, microwaves, terahertz waves, ultraviolet, infrared, x-rays, gamma rays, and / or any other light spectrum. This also includes 2D and 3D sensors. Examples of image sensor technologies can include CCD, CMOS, NMOS, etc. 3D sensors can be implemented using different technologies including stereo cameras, active stereo cameras, time of flight cameras, structured light cameras, radar, range imaging cameras, etc.

[0019] In embodiments of the presently disclosed subject matter, one or more of the steps illustrated in the drawings may be performed in a different order and / or one or more groups of steps may be performed simultaneously, or vice versa. The drawings are schematic diagrams of system architectures in accordance with embodiments of the presently disclosed subject matter. Each module in the drawings may be comprised of any combination of software, hardware, and / or firmware that performs the functions defined and described herein. The modules in the drawings may be consolidated in one location or distributed across one or more locations.

[0020] Some examples of the subject matter disclosed herein are not limited in application to the details of construction and arrangement of components set forth in the following description or in the drawings. The invention is capable of other embodiments and of being carried out in various ways. The terms and phrases employed herein are used for purposes of description and should not be regarded as limiting.

[0021] In this application, elements of a drawing that are not depicted within the drawing but are labeled with a number that appears in a previous drawing may have the same purpose and description as in the previous drawing.

[0022] The drawings in this application may not be to scale: different drawings may use different scales, and even within the same drawing, for example different views of the same object may be at different scales, or two adjacent objects may be at different scales.

[0023] FIG. 1A illustrates an exemplary communication system. In this example, the communication system can include one or more terminal devices 110 (terminal device 110A, terminal device 110B, etc.), one or more servers (server 120, etc.), a communication network 130, and one or more other computing devices (computing device 140, etc.). In this example, the terminal device 110 can communicate with the server 120 and the computing device 140. The communication can use the communication network 130 and / or can be direct communication. Possible implementations of the terminal device 110 and / or the server 120 and / or the computing device 140 can include the computing device 200 illustrated in FIG. 2A and FIG. 2B. In another example, the terminal device 110 and / or the server 120 and / or the computing device 140 can be implemented using a cloud platform, for example, as a virtual computing device or a physical computing device in the cloud platform. Other non-limiting examples of possible implementations of the terminal device 110 can include a mobile phone, a smartphone, a tablet, a wearable computing device, a personal computer (PC), etc. In some examples, terminal device 110 can communicate with server 120 and computing device 140. Possible implementations of terminal device 110 and / or server 120 and / or computing device 140 can connect to communication network 130 directly or through other communication devices using a local router. Non-limiting examples of communication network 130 can include the Internet, a telephone network, a cellular network, a satellite communication network, a private communication network, a virtual private network (VPN), and the like.

[0024] 1B illustrates an exemplary communication system. In this example, terminal device 110 communicates with server 120 using communication hardware 132. For example, terminal device 110 communicates with server 120 directly, through a different communication device, over a communication line, using a communication network (such as communication network 130). In this example, other computing devices 140 can observe and / or modify and / or interfere with the communication between terminal device 110 and server 120. In one example, the communication line between terminal device 110 and server 120 can be a direct communication line, a communication line through another device, and a communication line over a communication network (such as communication network 130).

[0025] 1A and 1B illustrate several possible implementations of a communication system. In some embodiments, other communication systems that enable communication between terminal device 110 and / or server 120 and / or computing device 140 can be used.

[0026] 2A is a simplified block diagram of one possible implementation of a computing device 200. In this example, computing device 200 may include one or more memory units 210, one or more processing units 220, and an input / output system 250. In some implementations, computing device 200 may include additional components or may exclude some of the components described above.

[0027] 2B is a block diagram of a possible implementation of a computing device 200. In this example, the computing device 200 may include one or more memory units 210, one or more processing units 220, one or more communication modules 230, one or more power sources 240, an input / output system 250, one or more image sensors 260, one or more light sources 265, one or more motion sensors 270, and one or more positioning sensors 275. In some implementations, the computing device 200 may include additional components, or may exclude some of the above-mentioned components. For example, in some implementations, the computing device 200 may include at least one of one or more user input devices, one or more output devices, etc. In other examples, in some implementations, at least one of the memory unit 210, the communication module 230, the power source 240, the input / output system 250, the image sensor 260, the light sources 265, the motion sensor 270, and the positioning sensor 275 may be excluded from the computing device 200.

[0028] In some embodiments, one or more power sources 240 are configured to provide power to computing device 200. Possible non-limiting implementations of power source 240 may include one or more batteries, one or more capacitors, one or more connections to external power sources, one or more power converters, and any combination thereof.

[0029] In some embodiments, the one or more processing units 220 are configured to execute software programs. For example, the processing unit 220 is configured to execute software programs stored in the memory unit 210. In some cases, the executed software programs can retrieve information from the memory unit 210. Possible non-limiting implementations of the processing unit 220 can include one or more single-core processors, one or more multi-core processors, one or more controllers, one or more application processors, one or more system-on-chip processors, one or more central processing units, one or more graphics processing units, one or more neural processing units, and any combination thereof.

[0030] In some embodiments, one or more of the communication modules 230 are configured to transmit and receive information. For example, control signals can be transmitted and / or received through the communication module 230. In another example, information received through the communication module 230 is stored in the memory unit 210. In an additional example, information retrieved from the memory unit 210 is transmitted using the communication module 230. In another example, input data can be transmitted and / or received using the communication module 230. Examples of such input data can include input data entered by a user using a user input device, information captured using one or more sensors, etc. Examples of such sensors can include audio sensors, image sensors 260, motion sensors 270, positioning sensors 275, temperature sensors, etc.

[0031] In some embodiments, the input / output system 250 can include or be configured to connect to input and / or output devices such as a keyboard, a computer mouse, a touchpad, a touch screen, a gesture recognition system, an image sensor, an audio sensor, a visual indicator (such as an LED visual indicator), an audio speaker, an audio pager, a display screen, an augmented reality system, etc. In some examples, the input and / or output devices are configured to generate digital signals based on input from a user, person, or environment, and / or to generate physical outputs to a user, person, or environment based on the digital signals. In one example, the input / output system 250 is configured to receive and / or store such digital signals from the memory unit 210. In one example, the input / output system 250 is configured to receive such digital signals from the processing unit 220 and / or the communication module 230, and / or to provide such digital signals to the processing unit 220 and / or the communication module 230.

[0032] In some embodiments, the one or more image sensors 260 are configured to capture visual information by converting light into image data, an image, a series of images, a video, a 3D image, a series of 3D images, a 3D video, or the like. In some examples, the captured visual information is stored in the memory unit 210. In some additional examples, the captured visual information is transmitted using the communication module 230, for example to another computer-controlled device. In some examples, the processing unit 220 can control the above-mentioned processes. For example, the processing unit 220 can control at least one of capturing the visual information, storing the captured visual information, transmitting the captured visual information, analyzing the captured visual information, and the like. In some cases, the captured visual information is processed by the processing unit 220. For example, the captured visual information is compressed by the processing unit 220, the compressed captured visual information is then stored in the memory unit 210, and the compressed captured visual information is then transmitted using the communication module 230. In other examples, the captured visual information is processed for object detection, event detection, motion detection, face detection, person detection, person recognition, and the like.

[0033] In some embodiments, one or more light sources 265 are configured to emit light, for example, to enable better image capture by image sensor 260. In some examples, the emission can be coordinated with the capture operation of image sensor 260. In some examples, the emission can be continuous. In some examples, the emission is performed at selected times. The emitted light can be visible light, infrared light, x-rays, gamma rays, and / or any other light spectrum. In some examples, image sensor 260 can capture the light emitted by light sources 265, for example, to capture 3D images and / or videos using active stereo techniques.

[0034] In some embodiments, the one or more motion sensors 270 are configured to perform at least one of: detect motion of objects in the environment of the computing device 200, measure a velocity of objects in the environment of the computing device 200, measure an acceleration of objects in the environment of the computing device 200, detect motion of the computing device 200, measure a velocity of the computing device 200, measure an acceleration of the computing device 200, etc. In some implementations, the one or more motion sensors 270 may include one or more accelerometers configured to measure a propensity acceleration of the computing device 200 and / or detect a change in a propensity acceleration. In some implementations, the one or more motion sensors 270 may include one or more gyroscopes configured to detect a change in an orientation of the computing device 200 and / or measure information related to an orientation of the computing device 200. In some implementations, the motion sensor 270 may be implemented using the image sensor 260, for example, by analyzing images captured by the image sensor 260 to perform at least one of the following tasks: tracking objects in the environment of the computing device 200, detecting moving objects in the environment of the computing device 200, measuring the velocity of objects in the environment of the computing device 200, measuring the acceleration of objects in the environment of the computing device 200, measuring the velocity of the computing device 200, for example, by calculating the ego-motion of the image sensor 260, measuring the acceleration of the computing device 200, for example, by calculating the ego-motion of the image sensor 260. In some implementations, the motion sensor 270 may be implemented using the image sensor 260 and the light source 265, for example, by implementing a LIDAR using the image sensor 260 and the light source 265. In some implementations, the motion sensor 270 may be implemented using one or more RADARs. In some examples, information captured using the motion sensor 270 can be stored in the memory unit 210, processed by the processing unit 220, and transmitted and / or received using the communication module 230.

[0035] In some embodiments, one or more positioning sensors 275 are configured to obtain positioning information of computing device 200 , detect changes in the position of computing device 200 , and / or measure the position of computing device 200 . In some examples, the positioning sensor 275 may be implemented using one of the following technologies: Global Positioning System (GPS), Global Navigation Satellite System (GLONASS), Galileo Global Navigation System, BeiDou navigation system, other Global Navigation Satellite System (GNSS), Indian Regional Navigation Satellite System (IRNSS), Local Positioning System (LPS), Real-Time Location System (RTLS), Indoor Positioning System (IPS), Wi-Fi based positioning system, cellular triangulation, etc. In some examples, information captured using the positioning sensor 275 can be stored in the memory unit 210 , processed by the processing unit 220 , and transmitted and / or received using the communication module 230 .

[0036] In some embodiments, methods such as methods 400, 420, 440, 460, 480, 500, 520, 540, 560, 700, 900, 920, and 940 may include one or more steps. In some examples, the methods, as well as all of the individual steps thereof, may be performed by various aspects, such as terminal device 110, server 120, computing device 140, computing device 200, etc. A system including at least one processor, such as processing unit 220, may perform any of the methods, as well as all of the individual steps thereof, such as by processing unit 220 executing software instructions stored in memory unit 210. In some examples, the methods, as well as all of the individual steps thereof, may be performed by dedicated hardware. In some examples, a computer-readable medium, such as a non-transitory computer-readable medium, may store computer-implemented instructions and / or data for performing any of the methods, as well as all of the individual steps thereof. Some non-limiting examples of possible ways of executing the method include continuous execution (e.g., returning to the beginning of the method when normal execution of the method is completed), periodic execution, executing the method at a selected time, execution upon detection of a trigger (some non-limiting examples of such triggers include a trigger from a user, a trigger from another process, a trigger from an external device, etc.), etc.

[0037] In some embodiments, a machine learning algorithm (also referred to in this disclosure as a machine learning model) is trained using training examples, such as those described below. Some non-limiting examples of such machine learning algorithms can include classification algorithms, data regression algorithms, image segmentation algorithms, visual detection algorithms (object detectors, human detectors, motion detectors, edge detectors, etc.), visual recognition algorithms (face recognition, human recognition, object recognition, etc.), speech recognition algorithms, mathematical embedding algorithms, natural language processing algorithms, support vector machines, random forests, nearest neighbor algorithms, deep learning algorithms, artificial neural network algorithms, convolutional neural network algorithms, regression neural network algorithms, linear machine learning models, nonlinear machine learning models, ensemble algorithms, etc. For example, the trained machine learning algorithms can include inference models such as predictive models, classification models, data regression models, clustering models, partitioning models, artificial neural networks (deep neural networks, convolutional neural networks, regression neural networks, etc.), random forests, support vector machines, etc. In some examples, the training examples can include example inputs along with desired outputs corresponding to the example inputs. Further, in some examples, training the machine learning algorithm using the training examples can generate a trained machine learning algorithm, and the trained machine learning algorithm is used to estimate outputs for inputs not included in the training examples. In some examples, the engineers, scientists, processes and machines that train the machine learning algorithm can further use validation examples and / or test examples. For example, the validation examples and / or test examples can include example inputs along with desired outputs corresponding to the example inputs, the trained machine learning algorithm and / or the moderately trained machine learning algorithm is used to estimate outputs for example inputs of the validation examples and / or test examples, the estimated outputs are compared to the desired outputs corresponding to the example inputs, and the trained machine learning algorithm and / or the moderately trained machine learning algorithm is evaluated based on the results of the comparison.In some examples, a machine learning algorithm can include parameters and hyperparameters, where the hyperparameters are set manually by a human or automatically by a process external to the machine learning algorithm (such as a hyperparameter search algorithm), and the parameters of the machine learning algorithm are set by the machine learning algorithm based on the training examples. In some implementations, the hyperparameters are set based on the training examples and the validation examples, and the parameters are set based on the training examples and the selected hyperparameters. For example, given the hyperparameters, the parameters can be conditionally independent of the validation examples.

[0038] In some embodiments, a trained machine learning algorithm (also referred to in this disclosure as a machine learning model and a trained machine learning model) is utilized to analyze an input and generate an output, for example, in the following cases: In some examples, the trained machine learning algorithm is utilized as an inference model that generates an inferred output when an input is provided. For example, the trained machine learning algorithm includes a classification algorithm, where the input includes samples, and the inferred output can include a classification of the samples (e.g., an inferred label, an inferred tag, etc.). In other examples, the trained machine learning algorithm includes a regression model, where the input includes samples, and the inferred output can include an estimate corresponding to the samples. In yet other examples, the trained machine learning algorithm includes a clustering model, where the input includes samples, and the inferred output can include an assignment of the samples to at least one cluster. In an additional example, the trained machine learning algorithm includes a classification algorithm, where the input includes an image, and the inferred output can include a classification of an item depicted in the image. In yet another example, the trained machine learning algorithm includes a regression model, where the input includes an image, and the inferred output can include an estimate corresponding to an item depicted in the image (e.g., an estimated characteristic of the item, such as dimensions, volume, age of a person depicted in the image, price of a product depicted in the image, etc.). In an additional example, the trained machine learning algorithm may include an image segmentation model, where the input includes an image and the inferred output may include segmentation of the image. In yet another example, the trained machine learning algorithm may include an object detector, where the input includes an image and the inferred output may include one or more detected objects in the image and / or one or more locations of the objects in the image.In some examples, the trained machine learning algorithm may include one or more formulas and / or one or more functions and / or one or more rules and / or one or more procedures, where the inputs are used as inputs to the formulas and / or functions and / or rules and / or procedures, and the inferred output may be based on the outputs of the formulas and / or functions and / or rules and / or procedures (e.g., selecting one of the outputs of the formulas and / or functions and / or rules and / or procedures, using a statistical measure of the outputs of the formulas and / or functions and / or rules and / or procedures, etc.).

[0039] In some embodiments, the artificial neural network is configured to analyze the input and generate a corresponding output in the cases described herein. Some non-limiting examples of such artificial neural networks can include shallow artificial neural networks, deep artificial neural networks, feedback artificial neural networks, feedforward artificial neural networks, autoencoder artificial neural networks, probabilistic artificial neural networks, time delay artificial neural networks, convolutional artificial neural networks, regression artificial neural networks, long short-term memory artificial neural networks, etc. In some examples, the artificial neural network is manually configured. For example, the structure of the artificial neural network is manually selected, the type of artificial neurons of the artificial neural network is manually selected, and the parameters of the artificial neural network (e.g., parameters of the artificial neurons of the artificial neural network) are manually selected. In some examples, the artificial neural network is configured using machine learning algorithms. For example, a user may select hyperparameters for an artificial neural network and / or a machine learning algorithm, and the machine learning algorithm may use the hyperparameters and training examples to determine the parameters of the artificial neural network, e.g., using backpropagation, using gradient descent, using stochastic gradient descent, and using mini-batch gradient descent. In some examples, an artificial neural network may be generated from two or more other artificial neural networks by combining the two or more other artificial neural networks into a single artificial neural network.

[0040] Some non-limiting examples of image data include images, grayscale images, color images, 2D images, 3D images, videos, 2D videos, 3D videos, frames, footages, data derived from other image data, and the like. In some embodiments, analyzing the image data (e.g., by the methods, steps, and modules described herein) can include analyzing the image data to obtain preprocessed image data, and then analyzing the image data and / or the preprocessed image data to obtain a desired result. The following are examples, and one skilled in the art can understand that the image data can be preprocessed using other types of preprocessing methods. In some examples, the image data can be preprocessed by transforming the image data using a transformation function to obtain transformed image data, and the preprocessed image data can include the transformed image data. For example, the transformed image data can include one or more convolutions of the image data. For example, the transformation function can include one or more image filters, such as a low pass filter, a high pass filter, a band pass filter, an all pass filter, and the like. In some examples, the transformation function can include a non-linear function. In some examples, the image data is preprocessed by smoothing at least a portion of the image data, for example, using a Gaussian convolution and using a median filter. In some examples, the image data is pre-processed to obtain different representations of the image data. For example, the pre-processed image data can include a representation of at least a portion of the image data in a frequency domain, a discrete Fourier transform of at least a portion of the image data, a discrete wavelet transform of at least a portion of the image data, a time / frequency representation of at least a portion of the image data, a representation of at least a portion of the image data in a lower dimension, a lossy representation of at least a portion of the image data, a lossless representation of at least a portion of the image data, a time-ordered series of any of the above, any combination of the above, etc. In some examples, the image data is pre-processed to extract edges, and the pre-processed image data can include information based on and / or related to the extracted edges. In some examples, the image data is pre-processed to extract image features from the image data.Some non-limiting examples of such image features can include information based on and / or related to edges, corners, blobs, ridges, Scale Invariant Feature Transform (SIFT) features, temporal features, etc. In some examples, analyzing the image data can include computing at least one convolution of at least a portion of the image data and utilizing the computed at least one convolution to compute at least one resultant value and / or to determine, identify, recognize, classify, etc.

[0041] In some embodiments, analyzing the image data (e.g., by methods, steps, and modules described herein) can include analyzing the image data and / or pre-processed image data utilizing one or more rules, functions, procedures, artificial neural networks, object detection algorithms, face detection algorithms, visual event detection algorithms, motion detection algorithms, motion detection algorithms, background removal algorithms, inference models, etc. Non-limiting examples of such inference models can include manually programmed inference models, classification models, regression models, results of training algorithms, such as machine learning algorithms and / or deep learning algorithms, using training examples, etc., where the training examples can include examples of data instances, and in some cases, the data instances can be labeled with corresponding desired labels and / or results. In some embodiments, analyzing the image data (e.g., by methods, steps, and modules described herein) can include analyzing pixels, voxels, point clouds, range data, etc., contained within the image data.

[0042] Convolution can include convolution of any dimension. A one-dimensional convolution is a function that transforms an original sequence into a transformed sequence. A one-dimensional convolution is defined by a sequence of scalars. Each particular value of the transformed sequence is determined by computing a linear combination of values ​​in a subsequence of the original sequence that corresponds to the particular value. The computed convolution result value can include any value of the transformed sequence. Similarly, an n-dimensional convolution is a function that transforms an original n-dimensional array into a transformed array. An n-dimensional convolution is defined by an n-dimensional array of scalars (known as the kernel of the n-dimensional convolution). Each particular value of the transformed array is determined by computing a linear combination of values ​​in an n-dimensional region of the original array that corresponds to the particular value. The computed convolution result value can include any value of the transformed array. In some examples, an image can include one or more components (such as color components, depth components, etc.), where each component can include a two-dimensional array of pixel values. In one example, computing the convolution of an image can include computing a two-dimensional convolution of one or more components of the image. In another example, computing the convolution of the image can include stacking arrays from different components to create a three-dimensional array and computing a three-dimensional convolution of the resulting three-dimensional array. In some examples, the video can include one or more components (e.g., color components, depth components, etc.), each component can include a three-dimensional array of pixel values ​​(having two spatial axes and one time axis). In one example, computing the convolution of the video can include computing a three-dimensional convolution of one or more components of the video. In another example, computing the convolution of the video can include stacking arrays from different components to create a four-dimensional array and computing a four-dimensional convolution of the resulting four-dimensional array. In some examples, the audio data can include one or more channels, each channel can include a stream or one-dimensional array of values. In one example, computing the convolution of the audio data can include computing a one-dimensional convolution of one or more channels of the audio data.In another example, computing a convolution of the audio data may include stacking arrays from different channels to create a two-dimensional array and computing a two-dimensional convolution of the resulting two-dimensional array.

[0043] In some examples, an object is removed from the visual content (e.g., by methods, steps, and modules described herein). In one example, segmentation data (such as a binary mask, an alpha matte, a probability mask, etc.) of an object in the visual content is obtained. For example, the segmentation data may be read from a memory, received from an external device, received from a user, or determined by analyzing the visual content (e.g., by an object segmentation algorithm). The segmentation data may then be utilized to remove the object from the visual content, and the area in the visual content where the object was previously depicted may be filled, for example, by stitching another object on top of it, for example, by using an inpainting algorithm, or by using an interpolation algorithm. In some examples, an object is inserted into the visual content (e.g., by methods, steps, and modules described herein). For example, a visual matting algorithm may be utilized to insert the object into the visual content. In another example, a visual stitching algorithm may be utilized to stitch the object to a selected location in the visual content. In some examples, the object may move within the visual content. For example, an object may be removed from an original location in the visual content (e.g., as described above) and inserted into a new location in the visual content (e.g., as described above) to create visual movement of the object from the original location to the new location. In some examples, the visual appearance of an object in the visual content may be modified. For example, an original representation of an object may be removed from the visual content (e.g., as described above) and a new representation of the object may be inserted into the visual content (e.g., as described above) to create a modification of the visual appearance of the object in the visual content.

[0044] Substantially visually similar is understood to include visually identical. For example, two substantially visually similar visual content can be the same visual content, and two substantially visually similar objects can be the same visual object. Substantially identical is understood to include identical. For example, two substantially identical orientations can be the same orientation.

[0045] In many cases, visual content utilized in training a machine learning generative model (e.g., as training examples, validation examples, and / or test examples) is associated with different resources (e.g., captured or generated using different tools from different contributors, belonging to different owners). For credit, it is important to determine the contribution of specific visual content, e.g., to understand the importance and / or efficiency of different resources for further training of the machine learning generative model or for obtaining and / or generating new effective visual content for training other machine learning models.

[0046] 3A and 3B illustrate exemplary visual content. FIG. 3A includes a plurality of visual contents 300 that are utilized to train a machine learning model, such as a generative machine learning model. In this example, the plurality of visual contents 300 can include visual content 302 of two people sitting around a table drinking coffee while talking, visual content 304 of a bottle, visual content 306 of a person standing, visual content 308 of a car, visual content 310 of different cars, visual content 312 of a bus, and visual content 314 of a tractor. In some examples, each visual content of the plurality of visual contents 300 is associated with a label, and a machine learning model is trained using the visual content 300 and the associated labels, for example, as training examples, validation examples, and / or test examples. For example, the labels can be discrete values ​​associated with classes (e.g., a binary value associated with one of two alternative classes, a ternary value associated with one of three alternative classes, etc.), and the machine learning model can be a machine learning classification model. In other examples, the labels can be numeric values, and the machine learning model can be a machine learning regression model. In some examples, each visual content of the plurality of visual contents 300 is associated with textual content, and a machine learning generative model that generates visual output from textual input is trained using the visual content 300 and the associated textual content, e.g., as training, validation, and / or test examples. Some non-limiting examples of such machine learning generative models include a Deep Generative Model (DGM), a Generative Adversarial Network model (GAN), an autoregressive model, a Variational AutoEncoder model (VAE), a Transformer-based generative model, etc. FIG. 3B illustrates an example visual content 350 generated by a generative model, e.g., a machine learning generative model trained using visual content 300.In this example, visual content 350 depicts two people sitting around a table talking and drinking coffee, with a bottle placed on the table.

[0047] 4A is a flow diagram of an example method 400 for attributing generated visual content to training cases. In this example, the method 400 can include receiving first visual content generated using a generative model, where the generative model is a result of training a machine learning algorithm using a plurality of training cases, each training case of the plurality of training cases being associated with visual content (step 402), determining one or more characteristics of the first visual content (step 404), for each training case of the plurality of training cases, analyzing the visual content associated with the training case and determining one or more characteristics of the visual content associated with the training case (step 406), attributing the first visual content to at least one, but not all, first subgroup of the plurality of training cases using the one or more characteristics of the first visual content and the characteristics of the visual content associated with the plurality of training cases (step 408), determining if at least one visual content associated with the training cases of the first subgroup is associated with at least one first resource (step 410), and, for each resource of the at least one first resource, updating a data record associated with the resource based on the attribution (step 412).

[0048] A non-limiting example of visual content generated using a generative model is visual content 350. In one example, visual content 350 can be visual content generated using a generative model, where the generative model is a result of training a machine learning algorithm using a plurality of training cases, and each visual content of the plurality of visual contents 300 is associated with each training case of the plurality of training cases. In this example, step 402 can receive visual content 350. Further, step 404 can determine one or more characteristics of visual content 350. For example, step 404 can analyze visual content 350 using an object detection algorithm to determine that two people are sitting around a table and a bottle is placed on the table (thereby determining characteristics of visual content 350). Further, step 406 can analyze each visual content of visual content 300 to determine one or more characteristics of the visual content. For example, step 406 may utilize an object detection algorithm to analyze visual content 302 and determine that visual content 302 depicts two people sitting around a table (thereby determining characteristics of visual content 302), and then analyze visual content 304 and determine that visual content 304 depicts a bottle (thereby determining characteristics of visual content 304). Further, step 408 may utilize one or more characteristics of visual content 350 (determined by step 404) and characteristics of visual content 300 (determined by step 406) to attribute visual content 350 to visual content 302 and visual content 304, but not to visual content 306, 308, 310, 312, and 314. Further, step 410 may determine that visual content 302 is associated with a first human contributor and that visual content 304 is an output of a first generative model (thereby associating visual content 304 with the first generative model).Further, step 412 may update the data record associated with the first person contributor, e.g., updating credits included in the data record and / or adding to the data record an indicator of visual content 350. Additionally or alternatively, step 412 may update the data record associated with the first generative model, e.g., updating credits included in the data record and / or adding to the data record an indicator of visual content 350.

[0049] In some examples, step 402 can include the first visual content being generated using a generative model. The generative model is a result of training a machine learning algorithm using a plurality of training examples, and each training example of the plurality of training examples is associated with visual content. Some non-limiting examples of visual content associated with such training examples are illustrated in FIG. 3A (e.g., visual content 302, visual content 304, etc.). In one example, the generative model can be a result of training a machine learning algorithm using a plurality of training examples and additional training examples. For example, receiving the first visual content according to step 402 can include retrieving the first visual content from a memory, receiving the first visual content from an external device (e.g., using a digital communication device), receiving the first visual content from a generative model, or generating the first visual content using a generative model. In one example, the first visual content is or can include at least one of an image, a video, an animation, an illustration, or a drawing. In one example, the first visual content can be or can include two-dimensional visual content. In one example, the first visual content can be or can include three-dimensional visual content. In one example, for each training case of the plurality of training cases, the visual content associated with the training case can be or can include at least one of an image, a video, an animation, an illustration, or a drawing. In one example, for each training case of the plurality of training cases, the visual content associated with the training case can be or can include two-dimensional visual content. In one example, for each training case of the plurality of training cases, the visual content associated with the training case can be or can include three-dimensional visual content.

[0050] In some examples, step 404 may include determining one or more characteristics of the first visual content received by step 402. In some examples, a machine learning model may be trained using training examples to determine image and / or video characteristics. Examples of such training examples may include sample images and / or video along with labels indicating one or more characteristics of the sample images and / or video. Step 404 may utilize the trained machine learning model to analyze the video content to determine one or more characteristics of the first visual content. In some examples, a data structure may associate visual content with characteristics, and step 404 may access the data structure based on the first visual content to determine one or more characteristics of the first visual content. In some examples, a function of pixel values ​​associated with the first visual content may be calculated to thereby determine one or more characteristics of the first visual content. Non-limiting examples of such functions may include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc. In some examples, step 404 may calculate a convolution of at least a portion of the first visual content, thereby obtaining a result value of the calculated convolution of at least a portion of the first visual content, and may determine one or more characteristics of the first visual content based on the result value of the calculated convolution of at least a portion of the first visual content. For example, when the result value is a first value, step 404 may determine a first value for the one or more characteristics, and when the result value is a second value, step 404 may determine a second value for the one or more characteristics. In some examples, step 404 may determine one or more characteristics of the first visual content based on intermediate results of a generative model in generating the first visual content. For example, the generative model may include multiple layers, and output values ​​of one layer may be used as inputs to a next layer, and the intermediate results may include at least a portion of these values.In some examples, step 404 can analyze the first visual content (e.g., utilizing a visual object detection algorithm) to detect at least a first object and a second object depicted in the first visual content, and can determine one or more characteristics of the first visual content based on a position of the first object in the first visual content and a position of the second object in the first visual content. For example, step 404 can determine one or more characteristics of the first visual content based on a distance between the objects, a relative orientation between the objects, etc. Non-limiting examples of such objects can include inanimate objects, animate objects, people, etc. Step 404 can analyze the first visual content (e.g., utilizing a visual object recognition algorithm) to determine a type of object depicted in the first visual content, and can determine one or more characteristics of the first visual content based on the type of object. For example, when the type of object is a first type, step 404 can determine a first value for the one or more characteristics, and when the type of object is a second type, step 404 can determine a second value for the one or more characteristics. The second value can be different from the first value. Non-limiting examples of such objects can include inanimate objects, animate objects, people, etc. Step 404 can analyze the first visual content to detect an event depicted in the first visual content (e.g., utilizing a visual event detection algorithm) and can determine one or more characteristics of the first visual content based on the event. For example, if an event is detected, step 404 can determine a first value for the one or more characteristics, and if an event is not detected, step 404 can determine a second value for the one or more characteristics. In another example, if an event is detected in a first portion of the first visual content, step 404 can determine a first value for the one or more characteristics, and if an event is detected in a second portion of the first visual content, step 404 can determine a second value for the one or more characteristics. The second value can be different from the first value.In some examples, step 404 can determine one or more characteristics of the first visual content based on a temporal relationship between the appearance of the first object and the appearance of the second object. For example, when the first object appears before the appearance of the second object, step 404 can determine a first value for the one or more characteristics, and when the first object appears after the appearance of the second object, step 404 can determine a second value for the one or more characteristics. In other examples, step 404 can determine a first value for the one or more characteristics when the time elapsed between the appearance of the first object and the appearance of the second object is below a selected threshold, and step 404 can determine a second value for the one or more characteristics when the time elapsed between the appearance of the first object and the appearance of the second object is above a selected threshold. The second value can be different from the first value. In some examples, the first visual content includes at least a first frame and a second frame, and step 404 may determine one or more characteristics of the first visual content based on a change in a spatial relationship of two objects appearing in the first visual content from the first frame to the second frame. For example, when the two objects move closer to each other, step 404 may determine a first value for the one or more characteristics, and when the two objects move further away from each other, step 404 may determine a second value for the one or more characteristics. The second value may be different from the first value.

[0051] In some examples, step 406 can, for each training case of the plurality of training cases of step 402, analyze visual content associated with the training case and determine one or more characteristics of the visual content associated with the training case, e.g., as described above for step 404 and the first visual content. In some examples, training the machine learning algorithm to generate the generative model can include an iterative process, where in each iteration of the iterative process, a training case of the plurality of training cases can be analyzed and a loss function can be updated, and step 406 can determine one or more characteristics of the visual content associated with the particular training case based on the update of the loss function in the iteration that includes the analysis of the particular training case. For example, when the update of the loss function is above a selected threshold, step 406 can select a first value for the one or more characteristics, and when the update of the loss function is below the selected threshold, step 406 can select a second value for the one or more characteristics. The second value can be different from the first value. In some examples, step 406 can determine one or more characteristics of the particular visual content associated with the particular training case of the plurality of training cases based on a convolution of at least a portion of the particular visual content. For example, step 406 may compute a convolution of at least a portion of the particular visual content, thereby obtaining a result value of the computed convolution of at least a portion of the particular visual content, and then determine one or more characteristics of the particular visual content based on the result value. In some examples, step 406 may determine one or more characteristics of the particular visual content associated with a particular training case of the multiple training cases based on a location of a first object in the particular visual content and a location of a second object in the particular visual content, as described above for step 404 and the first visual content. Step 406 may determine one or more characteristics of the particular visual content associated with a particular training case of the multiple training cases based on a type of object depicted in the particular visual content, as described above for step 404 and the first visual content.In some examples, step 406 can determine one or more characteristics of the particular visual content associated with a particular training case of the plurality of training cases based on an event depicted in the particular visual content, e.g., as described above with respect to step 404 and the first visual content. In some examples, step 406 can determine one or more characteristics of the particular visual content associated with a particular training case of the plurality of training cases based on a temporal relationship between an appearance of a first object and an appearance of a second object in the particular visual content, e.g., as described above with respect to step 404 and the first visual content. For example, when the first object appears before the second object appears, step 406 can determine a first value for the one or more characteristics, and when the first object appears after the second object appears, step 406 can determine a second value for the one or more characteristics. In other examples, step 406 can determine a first value for one or more characteristics when the time elapsed between the appearance of the first object and the appearance of the second object is below a selected threshold, and step 406 can determine a second value for one or more characteristics when the time elapsed between the appearance of the first object and the appearance of the second object is above a selected threshold. The second value can be different from the first value. In some examples, step 406 can determine one or more characteristics of a particular visual content associated with a particular training case of the plurality of training cases based on a change in a spatial relationship of two objects appearing within the particular visual content from a first frame of the particular visual content to a second frame of the particular visual content, e.g., as described above for step 404 and the first visual content. For example, step 406 can determine a first value for one or more characteristics when the two objects move closer to each other, and step 406 can determine a second value for one or more characteristics when the two objects move further away from each other. The second value can be different from the first value.

[0052] In some examples, step 408 can include attributing the first visual content to at least one, but not all, of the plurality of training cases using the one or more characteristics of the first visual content determined by step 404 and characteristics of visual content associated with the plurality of training cases determined by step 406. In one example, step 408 can select a first subgroup using method 460 using the one or more characteristics of the first visual content determined by step 404 and characteristics of visual content associated with the plurality of training cases determined by step 406. In one example, for each training case of the plurality of training cases, a similarity between the one or more characteristics of the first visual content and one or more characteristics of visual content associated with the training case is determined, e.g., using a similarity function, and step 408 can determine whether to include the training case in the first subgroup. For example, step 408 can include the training case in the first subgroup when the similarity is higher than a selected threshold, and step 408 can exclude the training case from the first subgroup when the similarity is lower than a selected threshold. Non-limiting examples of such similarity functions can include distance-based similarity functions, correlation-based similarity functions, linear similarity functions, non-linear similarity functions, exponential similarity functions, and the like. In one example, a classification machine learning model can be trained using the training cases to classify the training cases as attributing the visual content or not based on the characteristics. Examples of such training cases can include sample characteristics of the first sample visual content and sample characteristics of the second sample visual content along with a label indicating whether the first sample visual content is attributing the training case relative to the second sample visual content. Step 408 can utilize the trained classification machine learning model to determine, for each training case of the plurality of training cases, whether to include the training case in the first subgroup.In some examples, step 408 may further utilize one or more characteristics of the first visual content determined by step 404 and characteristics of the visual content associated with the first subgroup of step 408 (determined by step 406) to determine, for each training case of the first subgroup, a degree of membership of the first visual content to the training case. For example, a regression machine learning model may be trained using the training cases to determine a degree of membership of the visual content to the training case based on the characteristics. Examples of such training cases may include sample characteristics of the first sample visual content and sample characteristics of the second sample visual content along with a label indicating a degree of membership of the first sample visual content to the training case associated with the second sample visual content. Step 408 may utilize the trained regression machine learning model to analyze, for each training case of the first subgroup, the characteristics associated with the first visual content and the characteristics of the visual content associated with the training case to determine a degree of membership of the first visual content to the training case.

[0053] In some examples, step 404 can calculate a convolution of at least a portion of the first visual content and obtain a result value of the convolution of at least a portion of the first visual content, thereby obtaining a characteristic of the first visual content. Further, for each training case of the plurality of training cases of step 402, step 406 can calculate a convolution of at least a portion of the visual content associated with the training case and obtain a result value of the convolution of at least a portion of the visual content associated with the training case, thereby obtaining a characteristic of the visual content associated with the training case. Further, step 408 can utilize the result value of the convolution of at least a portion of the first visual content and the result value of the convolution of at least a portion of the visual content associated with the training case to attribute the first visual content to at least one, but not all, of the plurality of training cases.

[0054] In some examples, training the machine learning algorithm to generate the generative model in step 402 can include a first training step of obtaining an intermediate model using a first subgroup of the plurality of training examples in step 402, and a second training step of obtaining a generative model using a second subgroup of the plurality of training examples in step 402 and the intermediate model for initialization. In one example, the second subgroup can be different from the first subgroup. In one example, the first subgroup and the second subgroup can be the same. In one example, the first subgroup and the second subgroup can have, for example, no training examples in common or at least one training example in common. In one example, training the machine learning algorithm to generate the generative model in step 402 can include more than two steps. In some examples, step 408 can compare results associated with the first visual content and the intermediate model and results associated with the first visual content and the generative model. For example, the first visual content can be used as an input to an intermediate model to evaluate the intermediate model with respect to the first visual content to obtain an output of the intermediate model, thereby obtaining a result related to the first visual content and the intermediate model. Further, the first visual content can be used as an input to a generative model to evaluate the generative model with respect to the first visual content to obtain an output of the generative model, thereby obtaining a result related to the first visual content and the generative model. In some examples, for each training example of the second subgroup, step 408 can determine whether to attribute the first visual content to the training example based on a result of the comparison. For example, depending on the first result, step 408 can attribute the first visual content to the first training example and avoid attributing the first visual content to the second training example, and depending on the second result, step 408 can attribute the first visual content to the second training example and avoid attributing the first visual content to the first training example.In one example, responsive to the third outcome, step 408 can attribute the first visual content to both the first training example and the second training example. In one example, responsive to the fourth outcome, step 408 can avoid attributing the first visual content to both the first training example and the second training example.

[0055] In some examples, step 410 can include determining whether at least one visual content associated with the training cases of the first subgroup (from step 408) is associated with at least one first resource. For example, step 410 can access a data structure that associates training cases with resources based on the training cases of the first subgroup to identify a resource associated with the training cases of the first subgroup, thereby determining the at least one first resource. In other examples, step 410 can analyze at least one visual content associated with the training cases of the first subgroup to identify a resource associated with the training cases of the first subgroup, thereby determining the at least one first resource. For example, the visual content can include an artist's signature, and analysis of the visual content can detect the signature to identify the artist, thereby identifying the resource.

[0056] In some examples, step 412 may include, for each resource of the at least one first resource of step 410, updating a data record associated with the resource based on the attribution. For example, the data record may be stored in a memory (such as a digital memory), and step 412 may update data associated with the data record in the memory. In other examples, the data record may be maintained in an external device, and step 412 may send information to the external device (e.g., utilizing a digital communication device) to cause the external device to update the data record. In yet other examples, the data record may include data in a database, and step 412 may update the data in the database. In one example, step 412 may update the data record to record the attribution. In one example, step 412 may update the data record to include an encoding (e.g., a digital encoding) of information related to at least one of the attribution, the first visual content received by step 402, the one or more characteristics of the first visual content determined by step 404, the first subgroup of step 408, or the at least one visual content associated with the first subgroup of step 408. In some examples, for each resource of the at least one first resource, step 412 may further update the data record associated with the resource based on a degree of attribution associated with the resource, such as a degree of attribution associated with the resource determined by step 408. For example, the degree of attribution may be a number within a range of numbers, may be a discrete value, or may be a level (e.g., "high," "medium," "low," etc.). In one example, the degree of attribution associated with the resource is recorded in the data record. In another example, the value in the data record may be updated to a new value determined by calculating a function of the degree of attribution and the value associated with the resource. Non-limiting examples of such functions may include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, and the like.In some examples, the value in the data record may represent a credit associated with the resource, and step 412 may update the credit associated with the resource (and thus the value in the data record) based on the membership and / or degree of membership.

[0057] FIG. 4B is a flow diagram of an example method 420 for attributing generated visual content to training examples and selecting visual content for a particular application based on the attribution. In this example, method 420 may include receiving second visual content generated using the generative model (step 422), determining one or more characteristics of the second visual content (step 424), attributing the second visual content to at least one, but not all, of the multiple training cases using the one or more characteristics of the second visual content and characteristics of the visual content associated with the multiple training cases, where the second subgroup may include at least one training case not included in the first subgroup (step 426), determining whether the at least one visual content associated with the training cases of the second subgroup is associated with at least one second resource, where the at least one second resource may include one or more resources not included in the at least one first resource (step 428), refraining from using the second visual content based on the at least one second resource (step 430), and commencing using the first visual content (step 432).

[0058] In some examples, step 422 may include receiving second visual content generated using a generative model (e.g., the generative model of step 402, a generative model different from the generative model of step 402 that is trained using the plurality of training examples of step 402, etc.), e.g., as described above with respect to step 402 and the first visual content. In some examples, step 424 may include determining one or more characteristics of the second visual content received by step 422, e.g., as described above with respect to step 404 and the first visual content. In some examples, step 426 may include attributing the second visual content to at least one but not all of the second subgroup of the plurality of training examples of step 402 using one or more characteristics of the second visual content determined by step 424 and characteristics of the visual content associated with the plurality of training examples (determined by step 406), e.g., as described above with respect to step 408, the first visual content, and at least one but not all of the first subgroup of the plurality of training examples. In one example, the second subgroup can include at least one training example not included in the first subgroup. In one example, the first subgroup and the second subgroup can, for example, have no common training examples or have at least one common training example. In one example, the first subgroup can be identical to the second subgroup. In some examples, step 428 can include determining whether at least one visual content associated with the training examples of the second subgroup is associated with at least one second resource, for example, as described above for step 410, the first visual content, and the at least one first resource. In one example, the at least one second resource can include one or more resources not included in the at least one first resource. In one example, the at least one first resource and the at least one second resource can, for example, have no common resource or have at least one common resource.In one example, the at least one second resource can be identical to the at least one first resource.

[0059] In some examples, step 430 can include refraining from utilizing the second visual content, for example, based on the at least one second resource. For example, refraining from utilizing can include avoiding storing certain digital information when stored would cause utilization of the second visual content by other processes. In other examples, refraining from utilizing can include avoiding transmission of certain digital signals to an external device, which when transmitted to the external device would cause the external device to utilize the second visual content. In yet other examples, refraining from utilizing can include avoiding presentation of the second visual content. In additional examples, refraining from utilizing can include avoiding insertion of the second visual content into the publication. In one example, step 430 can refrain from utilizing the second visual content in response to one or more of the at least one second resource being included in the selected resource group. In additional examples, step 430 can refrain from utilizing the second visual content in response to a cost associated with the second visual content based on the at least one second resource. In yet another example, step 430 can refrain from utilizing the second visual content in response to a cost associated with the second visual content based on the at least one second resource being higher than a cost associated with the first visual content based on the at least one first resource. For example, step 430 can determine a cost associated with the first visual content based on the at least one first resource, can determine a cost associated with the second visual content based on the at least one second resource, and can compare the determined cost associated with the first visual content and the determined cost associated with the second visual content. In one example, step 430 can decide to refrain from utilizing the second visual content based on a result of the comparison. In one example, step 432 can decide to commence utilizing the first visual content based on a result of the comparison.

[0060] In some examples, step 432 can include initiating utilization of the first visual content. For example, the digital information can be stored in a memory, causing utilization of the first visual content by another process. In other examples, a digital signal can be sent to an external device (e.g., using a digital communication device), causing the external device to utilize the first visual content. In yet other examples, the first visual content can be visually presented to an individual, for example, via a user interface. In additional examples, the first visual content can be included in a publication, included in a website, or included in a user interface, for example.

[0061] 4C is a flow diagram of an example method 440 for attributing generated visual content to training examples and selecting visual content for a particular application based on attribution. In this example, the method 440 includes receiving second visual content generated using a generative model (step 422), determining one or more characteristics of the second visual content (step 424), attributing the second visual content to at least one, but not all, of the plurality of training examples using the one or more characteristics of the second visual content and characteristics of the visual content associated with the plurality of training examples, the second subgroup may include at least one training example that is not included in the first subgroup (step 426), and accessing a data structure relating visual content to quantities (step 440). 8), utilizing the data structure to determine whether at least one visual content associated with the training examples of a first subgroup is associated with a first total amount (step 450), utilizing the data structure to determine whether at least one visual content associated with the training examples of a second subgroup is associated with a second total amount (step 452), refraining from use of the second visual content and beginning use of the first visual content based on the first and second total amounts (step 454), and updating data records associated with the resource according to step 412 based on the first total amount (step 456).

[0062] In some examples, step 448 can include accessing a data structure that associates visual content with quantities. For example, accessing a data structure that associates visual content with quantities can include accessing a memory that includes the data structure, can include communicating with an external device that holds the data structure, or can include accessing a database that implements the data structure.

[0063] In some examples, step 450 may include utilizing the data structure accessed by step 448 to determine whether at least one visual content associated with the training examples of the first subgroup of step 408 is associated with a first amount. For example, the first amount may be a amount associated with at least one visual content in the data structure, may be a sum of the amounts associated with at least one visual content in the data structure, or may be a function (such as a linear function, a nonlinear function, a polynomial function, an exponential function, etc.) of the amounts associated with at least one visual content in the data structure.

[0064] In some examples, step 452 may include determining whether at least one visual content associated with the training examples of the second subgroup is associated with a second total amount, e.g., utilizing a data structure as described above for step 450, the at least one visual content associated with the training examples of the first subgroup, and the first total amount.

[0065] In some examples, step 454 may include refraining from utilizing the second visual content (e.g., as described above for step 430) and beginning to utilize the first visual content (e.g., as described above for step 432), e.g., based on the first and second total amounts determined by steps 450 and 452.

[0066] In some examples, step 456 may further include updating a data record associated with the resource according to step 412 based on the first amount. For example, the first amount is recorded in at least one data record. In other examples, a value in the data record is updated to a new value determined by calculating a function of the first amount and the value. Non-limiting examples of such functions may include a linear function, a non-linear function, an exponential function, a polynomial function, a logarithmic function, etc.

[0067] 4D is a flow diagram of an example method 460 for attributing generated visual content to training cases. In this example, method 460 may include embedding the first visual content into a mathematical space using one or more characteristics of the first visual content (step 462), for each training case of the plurality of training cases, embedding the visual content associated with the training case into the mathematical space using one or more characteristics of the visual content associated with the training case (step 464), and selecting a first subgroup of at least one, but not all, of the plurality of training cases using the mathematical space (step 466).

[0068] In some examples, step 462 may include embedding the first visual content in the mathematical space utilizing one or more characteristics of the first visual content determined by step 404. For example, a function of the first visual content is calculated to determine a mathematical object associated with the first visual content in the mathematical space. For example, the function may be a function of pixel values ​​of the first visual content, may be a function of voxels of the first visual content, may be a function of low-level features of the first visual content, or may be a function of one or more characteristics of the first visual content determined by step 404. Non-limiting examples of such functions may include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc. In other examples, each mathematical object in the mathematical space may correspond to visual content, and the mathematical object corresponding to the first visual content may be an embedding of the first visual content in the mathematical space. The mathematical object may be determined, for example, based on a mapping or utilizing a data structure that associates mathematical objects with visual content.

[0069] In some examples, step 464 may include, for each training case of the plurality of training cases, embedding visual content associated with the training case within a mathematical space using one or more characteristics of the visual content associated with the training case determined by step 406, e.g., as described above for step 462 and the first visual content.

[0070] In some examples, step 466 may include selecting a first subgroup of at least one, but not all, of the plurality of training examples using the mathematical space. For example, one or more regions of the mathematical space are selected based on a mathematical object corresponding to the first visual content determined by step 462. Non-limiting examples of such regions may include a manifold, a subspace, a geometric region, and the like. In one example, the selected region may be, for example, a spherical region centered on the mathematical object corresponding to the first visual content, or a convex hull of the mathematical object corresponding to the first visual content and two or more other mathematical objects. In one example, the mathematical object corresponding to the first visual content may be the region of the mathematical space itself, and the selected region may be the mathematical object. In one example, step 466 may include in the first subgroup all training examples associated with visual content corresponding to the mathematical object (determined by step 464) within the selected first region of the mathematical space. In another example, step 466 can exclude from the first subgroup all training examples associated with visual content that corresponds to mathematical objects (as determined by step 464) within the selected second region of mathematical space.

[0071] 4E is a flow diagram of an example method 480 for attributing generated visual content to training cases. In this example, method 480 can include attributing parameters of the generative model to at least one, but not all, second subgroup of the plurality of training cases using parameters of the generative model and characteristics of visual content associated with the plurality of training cases (step 482), determining if at least one visual content associated with the training cases of the second subgroup is associated with at least one second resource (step 484), and, for each of the at least one second resource, updating a data record associated with the resource based on the attribution of the parameters of the generative model (step 486).

[0072] In some examples, step 482 can include utilizing the parameters of the generative model of step 402 and characteristics of the visual content associated with the plurality of training cases determined by step 406 to attribute the parameters of the generative model to a second subgroup of at least one, but not all, of the plurality of training cases. In one example, the second subgroup can be identical to the first subgroup. In another example, the second subgroup can be different from the first subgroup. In one example, a classification machine learning model can be trained using the training cases to classify whether the parameters of the generative model are attributed to the visual content or based on characteristics of the visual content. Examples of such training cases can include sample values ​​of sample parameters of the sample generative model and sample characteristics of the sample visual content along with a label indicating whether the sample parameters are attributed to a training case associated with the sample visual content. Step 482 can utilize the trained classification machine learning model to determine, for each training case of the plurality of training cases, whether the training case should be included in the second subgroup. In some examples, step 482 may further utilize the parameters of the generative model of step 402 and characteristics (determined by step 406) of the visual content associated with the second subgroup of step 482 to determine, for each training case of the second subgroup, a degree of attribution of the parameters to the training case. For example, a regression machine learning model may be trained using the training cases to determine a degree of attribution of the parameters of the generative model to the training cases based on characteristics of the visual content associated with the training cases. Examples of such training cases may include sample values ​​of sample parameters of the sample generative model and sample characteristics of the second sample visual content, along with labels indicating a degree of attribution of the sample parameters to the training cases associated with the second sample visual content. Step 482 may utilize the trained regression machine learning model to analyze, for each training case of the second subgroup, the parameters of the generative model of step 402 and characteristics of the visual content associated with the training cases to determine a degree of attribution of the parameters to the training cases.

[0073] In some examples, step 484 may include determining whether at least one visual content associated with the training examples of the second subgroup (of step 482) is associated with at least one second resource, e.g., as described above for step 410, the first subgroup, and the at least one first resource. In some examples, step 486 may include, for each resource of the at least one second resource, updating a data record associated with the resource based on an attribution of parameters of the generative model, e.g., as described above for step 412 and the at least one first resource. In one example, step 486 may update the data record to record the attribution. In one example, step 486 may update the data record to include an encoding (e.g., a digital encoding) of information associated with at least one of the attributions, parameters, the generative model of step 402, the second subgroup of step 482, or the at least one visual content associated with the second subgroup of step 482. In some examples, for each resource of the at least one first resource, step 486 may further update a data record associated with the resource based on a degree of attribution associated with the resource, such as the degree of attribution associated with the resource determined by step 484. For example, the degree of attribution may be a number within a range of numbers, may be a discrete value, or may be a level (such as "high", "medium", "low", etc.). In one example, the degree of attribution associated with the resource is recorded in the data record. In another example, the value in the data record may be updated to a new value determined by calculating a function of the degree of attribution and the value associated with the resource. Non-limiting examples of such functions may include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc. In some examples, the value in the data record may represent a credit associated with the resource, and step 486 may update the credit associated with the resource (and thus the value in the data record) based on the attribution and / or the degree of attribution.

[0074] An inference model can be, for example, the result of training a machine learning algorithm using multiple training examples, as described above. In many cases, an inference model can be distributed and / or used without disclosing at least some of the training examples used to train the machine learning algorithm. For example, it may be important to determine whether specific visual content was used to train the machine learning algorithm, for example, for credit, to obtain and / or generate new effective visual content for further training of the machine learning model, or to understand the importance and / or efficiency of different visual content for training other machine learning models.

[0075] 5A is a flow diagram of an example method 500 for identifying visual content used to train an inference model. In this example, method 500 can include receiving distinctive visual content (step 502), accessing data based on at least one parameter of an inference model, the inference model being a result of training a machine learning algorithm using a plurality of training cases, each training case of the plurality of training cases can include visual content (step 504), analyzing the data and the distinctive visual content to determine a likelihood that the distinctive visual content is included in at least one training case of the plurality of training cases (step 506), and generating a digital signal indicative of the likelihood that the distinctive visual content is included in at least one training case of the plurality of training cases (step 508).

[0076] In one example, step 502 can receive the visual content of FIG. 6C and step 504 can receive data based on at least one parameter of a generative model trained using visual content 300 as a training set without receiving direct access to visual content 300. Further, step 506 can analyze the visual content and data of FIG. 6C to determine a likelihood that the visual content of FIG. 6C is included in at least one training set of visual content 300. Further, step 508 can generate a digital signal indicative of a determination of a likelihood that the visual content of FIG. 6C is included in at least one training set of visual content 300. In another example, step 502 can receive the visual content of FIG. 6A and step 504 can receive data based on at least one parameter of a generative model trained using visual content 300 as a training set without receiving direct access to visual content 300. Further, step 506 can analyze the visual content and data of FIG. 6A to determine a likelihood that the visual content of FIG. 6A is not included in at least one training set of visual content 300. Additionally, step 508 may generate a digital signal indicative of a determination of the likelihood that the visual content of FIG. 6A is not included in at least one training example of visual content 300.

[0077] In some examples, step 502 may include receiving the unique visual content. For example, receiving the unique visual content by step 502 may include retrieving the unique visual content from a memory, receiving the unique visual content from an external device (e.g., using a digital communication device), receiving the unique visual content from a generative model, or generating the unique visual content using a generative model. In one example, the unique visual content may include at least one of an image, a video, an animation, an illustration, or a drawing. In one example, the unique visual content may be or may include two-dimensional visual content. In one example, the unique visual content may be or may include three-dimensional visual content. In one example, step 502 may access a stock photo repository to obtain the unique visual content. In one example, the unique visual content may include at least a portion of a frame of a video. In one example, the unique visual content may include a plurality of frames of a video. In one example, the unique visual content may include a still image.

[0078] In some examples, step 504 can include accessing data based on at least one parameter of the inference model. The inference model can be a result of training a machine learning algorithm using a plurality of training cases. Each training case of the plurality of training cases can include visual content. For example, accessing data based on at least one parameter of the inference model using step 504 can include accessing data in a digital memory, can include communicating with an external device that holds the data (e.g., using a digital communication device), can include determining the data using the at least one parameter, or can include determining the data using the inference model (thereby using the at least one parameter). In one example, step 504 can determine the data using method 520. In one example, the data based on at least one parameter of the inference model can be a value of the parameter. In another example, the data based on at least one parameter of the inference model can be a function of a value of the parameter. Non-limiting examples of such functions can include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc. In yet another example, the data based on the at least one parameter of the inference model can be one or more outputs of the inference model or can be based on one or more outputs of the inference model. In one example, the at least one parameter of the inference model can include at least one threshold of the inference model. In one example, the at least one parameter of the inference model can include at least a weight of a regularization term of the inference model. In one example, the at least one parameter of the inference model can include a weight of the inference model. In one example, the at least one parameter of the inference model can include at least one of a number of layers of the inference model, a type of at least one neuron of the inference model, a representation of the structure of the inference model (e.g., in a graph format), a type of rule of the inference model, or a width of at least one layer of the inference model. In some examples, the inference model can be a generative model.Further, for each training instance of the plurality of training instances, the visual content included in the training instance can be or include a pair of visual content, an input visual content, and a desired output visual content. In another example, for each training instance of the plurality of training instances, the visual content included in the training instance can be or include at least one of an input visual content and a desired output visual content.

[0079] In some examples, step 504 may utilize the unique visual content received by step 502 and at least one parameter of the inference model to obtain an output of the inference model corresponding to the unique visual content. For example, the inference model may be evaluated (utilizing the at least one parameter) with the unique visual content used as an input to the inference model to obtain an output. In some examples, step 504 may access data based on the output of the inference model corresponding to the unique visual content. For example, when the obtained output is a first output, step 504 may select a first value for the data, and when the obtained output is a second output, step 504 may select a second value for the data. In other examples, the data may be or may include the obtained output. In yet other examples, the data may be a function of the obtained output. Non-limiting examples of such functions may include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc.

[0080] In some examples, step 504 may utilize the characteristic visual content received by step 502 and at least one parameter of the inference model to obtain a gradient corresponding to the characteristic visual content. For example, a function may be selected based on the at least one parameter (e.g., the at least one parameter may be a parameter that selects a function of a family of functions) and a gradient of the selected function of the characteristic visual content may be calculated or estimated. In one example, the selected function may be an inference model. In some examples, step 504 may access data based on the gradient corresponding to the characteristic visual content. For example, when the gradient is a first gradient, step 504 may select a first value for the data, and when the gradient is a second gradient, step 504 may select a second value for the data. In other examples, the data may be or may include a gradient. In yet other examples, the data may be a function of the gradient. Non-limiting examples of such functions may include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc.

[0081] In some examples, step 504 can utilize the distinctive visual content and at least one parameter of the inference model received by step 502 to calculate a loss function associated with the machine learning algorithm and a loss corresponding to the distinctive visual content. For example, the inference model can be estimated (utilizing the at least one parameter) with the distinctive visual content utilized as an input of the inference model to obtain an output of the inference model, and the output of the inference model can be utilized as an input of the loss function to estimate the loss function, thereby calculating the loss. In some examples, step 504 can access data based on the loss function and the loss corresponding to the distinctive visual content. For example, when the loss is a first loss, step 504 can select a first value for the data, and when the loss is a second loss, step 504 can select a second value for the data. In other examples, the data can be or include a loss. In yet other examples, the data can be a function of the loss. Non-limiting examples of such functions can include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, and the like.

[0082] In some examples, step 506 can analyze the visual content received by step 502 and the data accessed by step 504 to determine a likelihood that the distinctive visual content is included in at least one training case of the plurality of training cases of step 504. For example, the likelihood can be a binary likelihood, a likelihood within a selected range of likelihood (e.g., a continuous range of likelihood), or a level of likelihood (e.g., “high”, “medium”, “low”, etc.). In one example, the likelihood can be a binary likelihood, and determining the likelihood that the distinctive visual content is included in at least one training case of the plurality of training cases by step 506 can include determining whether the distinctive visual content is included in at least one training case of the plurality of training cases. In one example, a classification machine learning model can be trained using the training cases and classify whether the visual content is included in a training case associated with an inference model based on the characteristics and the data. Such examples of training cases can include example data associated with example characteristics of the sample visual content and example parameters of the sample inference model along with a label indicating whether the sample visual content is associated with a training case used to train the machine learning algorithm to obtain the sample inference model. Step 506 may utilize the trained classification machine learning model to analyze the distinctive visual content received by step 502 and the data accessed by step 504 to determine a binary likelihood that the distinctive visual content is included in at least one training example of the plurality of training examples of step 504 (i.e., it may be determined that the distinctive visual content is or is not included in at least one training example of the plurality of training examples). In one example, a regression machine learning model may be trained using the training examples to determine the likelihood that visual content is included in the training examples associated with the inference model based on the characteristics.Such example training cases may include sample data associated with sample characteristics of the sample visual content and sample parameters of the sample inference model, along with a label indicating the likelihood that the sample visual content is associated with a training case used to train the machine learning algorithm to obtain the sample inference model. Step 506 may utilize the trained regression machine learning model to analyze the unique visual content received by step 502 and the data accessed by step 504 to determine a likelihood that the unique visual content is included in at least one training case of the multiple training cases of step 504.

[0083] In some examples, step 508 may include generating a digital signal indicative of a likelihood that the distinctive visual content (as determined by step 506) is included in at least one training example of the plurality of training examples. For example, the digital signal may include encoding of the likelihood. In other examples, step 508 may calculate a function of the likelihood to obtain an output value, and the digital signal may include the output value. Non-limiting examples of such functions may include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, and the like. In one example, the digital signal may be transmitted to an external device utilizing a digital communication device. In another example, the digital signal may cause storage of the likelihood or likelihood-based information in a digital memory. In yet another example, the digital signal may cause presentation of the likelihood or likelihood-based information (e.g., visual presentation, audible presentation, graphical presentation, textual presentation, etc. via a user interface). In some examples, a data structure associating visual content with resources may be accessed to determine if the distinctive visual content is associated with the distinctive resource, as described above with respect to step 410. Additionally, step 508 may include an indication of the distinctive resource in the generated digital signal.

[0084] Additionally or alternatively, the data accessed by step 504 and the distinctive visual content received by step 502 may be analyzed to determine a likelihood that an enhanced version of the distinctive visual content is included in at least one training example of the plurality of training examples. For example, the enhanced version of the particular visual content may be any visual content obtained by recursively applying an enhancement function of a selected group of enhancement functions to the particular visual content up to a selected number of times. In another example, the enhanced version of the particular visual content may be any visual content that is similar to the particular visual content based on a selected similarity function (e.g., a visual content may be similar to the particular visual content when the similarity between the particular visual content and the particular visual content is above a selected threshold). In yet another example, the enhanced version of the particular visual content may be any visual content that is less than the distance from the particular visual content selected based on a selected dissimilarity function. In one example, the likelihood may be a binary likelihood, a likelihood within a selected range of likelihoods (e.g., a continuous range of likelihoods), or a level of likelihood (e.g., "high", "medium", "low"). In one example, the likelihood can be a binary likelihood, and determining the likelihood that the enhanced version of the distinctive visual content is included in at least one training example of the plurality of training examples can include determining whether the enhanced version of the distinctive visual content is included in at least one training example of the plurality of training examples. In one example, the classification machine learning model can be trained using the training examples and classify whether the enhanced version of the visual content is included in a training example associated with an inference model based on the characteristics and the data. Examples of such training examples can include example data associated with example characteristics of the sample visual content and example parameters of the sample inference model, along with a label indicating whether the enhanced version of the sample visual content is associated with a training example used to train the machine learning algorithm to obtain the sample inference model.The trained classification machine learning model may be utilized to analyze the distinctive visual content received by step 502 and the data accessed by step 504 to determine a binary likelihood that the enhanced version of the distinctive visual content is included in at least one training example of the plurality of training examples of step 504 (i.e., it may be determined that the enhanced version of the distinctive visual content is or is not included in at least one training example of the plurality of training examples). In one example, a regression machine learning model may be trained using the training examples to determine the likelihood that the enhanced version of the visual content is included in a training example associated with an inference model based on the characteristics. Examples of such training examples may include sample data associated with sample characteristics of the sample visual content and sample parameters of the sample inference model along with a label indicating the likelihood that the enhanced version of the sample visual content is associated with a training example utilized in training the machine learning algorithm to obtain the sample inference model. The trained regression machine learning model may be utilized to analyze the distinctive visual content received by step 502 and the data accessed by step 504 to determine the likelihood that the enhanced version of the distinctive visual content is included in at least one training example of the plurality of training examples of step 504. In some examples, a second digital signal is generated that indicates a likelihood that the augmented version of the distinctive visual content is included in at least one training example of the plurality of training examples. For example, the second digital signal can include an encoding of the likelihood. In other examples, a function of the likelihood is calculated to obtain an output value, and the second digital signal can include the output value. Non-limiting examples of such functions can include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc. In one example, the second digital signal is transmitted to an external device using a digital communication device. In other examples, the second digital signal can cause storage of the likelihood or information based on the likelihood in a digital memory.In yet other examples, the second digital signal can cause a presentation of the likelihood or information based on the likelihood (e.g., a visual presentation, an audible presentation, a graphic presentation, a textual presentation, etc. via a user interface).

[0085] Additionally or alternatively, the data accessed by step 504 and the distinctive visual content received by step 502 may be analyzed to determine a similarity of the distinctive visual content to a selected training case of the plurality of training cases of step 504. Dissimilarity is also understood to be a similarity. Distance is also understood to be a similarity. Correlation is also understood to be a similarity. In one example, the similarity may indicate an amount of extension. In one example, the selected training case of the plurality of training cases may be a training case that is most similar and / or closest to the distinctive visual content in the plurality of training cases. In another example, the selected training case may be a training case associated with a distinctive resource. In yet another example, the selected training case may be a training case associated with an impact on the inference model above a selected threshold. For example, the impact may be determined based on an amount of change of the training case caused to an intermediate inference model during training. In one example, a regression machine learning model may be trained using the training cases to determine a similarity between the visual content and the training cases used to train the inference model. Such training example examples may include sample data associated with the sample visual content and the sample inference model, along with labels indicative of the similarity of the sample visual content and the selected training example used to train the inference model. A regression machine learning model may be utilized to analyze the unique visual content received by step 502 and the data accessed by step 504 to determine the similarity of the unique visual content to the selected training example of the plurality of training examples of step 504. In some examples, a second digital signal is generated indicative of the similarity of the unique visual content to the selected training example of the plurality of training examples. For example, the second digital signal may include an encoding of the similarity. In other examples, a function of the similarity is calculated to obtain an output value, and the second digital signal may include the output value. Non-limiting examples of such functions may include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc. In yet other examples, the second digital signal may be indicative of the selected training example of the plurality of training examples.In one example, the second digital signal is transmitted to an external device utilizing a digital communication device. In another example, the second digital signal can cause storage of the similarity or similarity-based information in a digital memory. In yet another example, the second digital signal can cause presentation of the similarity or similarity-based information (e.g., visual presentation, audible presentation, graphic presentation, textual presentation, etc. via a user interface).

[0086] 5B is a flow diagram of an example method 520 for identifying visual content utilized in training an inference model. In this example, method 520 may include utilizing the distinctive visual content to generate a plurality of variants of the distinctive visual content (step 522), for each variant of the plurality of variants, obtaining an output of the inference model corresponding to the variant using the variant and at least one parameter of the inference model (step 524), and based on the output of the inference model corresponding to the plurality of variants (step 526).

[0087] In some examples, step 522 may include generating a plurality of variants of the characteristic visual content using the characteristic visual content received by step 502. In one example, step 522 may generate a plurality of variants of the characteristic visual content using step 540. In one example, an augmentation technique may be applied to obtain the variant of the characteristic visual content from the characteristic visual content. Non-limiting examples of such augmentation techniques may include rotation, shear, skew, flip, noise, blur, adding objects, removing objects, cropping, frame reverse, upsampling, downsampling, frame mirroring, frame skipping, augmentation using a generative model to modify the original visual content, and the like. In one example, the augmentation technique may include any one of the above techniques. In other examples, the augmentation technique may include two or more of the above techniques, applied one after the other in a selected order. In some examples, step 522 may select a plurality of objects, and none of the plurality of objects are depicted in the characteristic visual content received by step 502. For example, step 522 may analyze the distinctive visual content received by step 502 utilizing an object recognition algorithm to determine objects depicted in the distinctive visual content, and may select the plurality of objects to include objects not included in the distinctive visual content of the group of three or more alternative objects. Additionally, step 522 may analyze the distinctive visual content received by step 502 for each object of the plurality of objects, and generate a variant of the distinctive visual content that includes a depiction of the object (e.g., utilizing a visual object stitching algorithm or utilizing a visual matting algorithm), thereby generating a plurality of variants of the distinctive visual content. In some examples, step 522 may analyze the distinctive visual content received by step 502, and detect a plurality of objects depicted in the distinctive visual content, for example, utilizing an object detection algorithm.Additionally, step 522 may analyze the distinctive visual content for each object of the plurality of objects and generate a variant of the distinctive visual content that does not include a depiction of the object, thereby generating a plurality of variants of the distinctive visual content. For example, pixels depicting the object in the distinctive visual content may be identified using a semantic segmentation algorithm, the pixels depicting the object may be erased, and new values ​​for the erased pixels may be determined using an inpainting algorithm.

[0088] In some examples, step 524 may, for each variant of the plurality of variants (of step 522), utilize the variant and at least one parameter of the inference model of step 504 to obtain an output of the inference model corresponding to the variant. For example, the inference model may be evaluated (using the at least one parameter) with the variant as an input to obtain an output of the inference model corresponding to the variant.

[0089] In some examples, step 526 can include the data of step 504 being based on the output of the inference model (obtained by step 524) corresponding to the multiple variations. For example, if the output obtained by step 524 is a first output, step 526 can select a first value for the data of step 504, and if the output obtained by step 524 is a second output, step 526 can select a second value for the data of step 504. In other examples, the data of step 504 can be or include the output obtained by step 524. In yet other examples, the data of step 504 can be a function of the output obtained by step 524. Non-limiting examples of such functions can include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, and the like. In additional examples, a machine learning model is trained to generate data based on the output. Examples of such training cases can include sample outputs along with labels indicating the desired data generated for the sample outputs. Step 526 can utilize the trained machine learning model to analyze the output obtained by step 524 to generate the data of step 504.

[0090] 5C is a flow diagram of an example method 540 for identifying visual content utilized in training an inference model. In this example, the method 540 can include obtaining a plurality of directions in a mathematical space (step 542), obtaining a unique mathematical object in the mathematical space corresponding to the unique visual content (step 544), for each of the plurality of directions, using the unique mathematical object and direction to determine a mathematical object in the mathematical space corresponding to the unique mathematical object and direction (step 546), and generating visual content corresponding to the mathematical object in the mathematical space corresponding to the unique mathematical object and direction for each of the plurality of directions, thereby generating a plurality of variations of the unique visual content (step 548). In one example, the mathematical space can be a non-orientable space. Non-limiting examples of such non-orientable spaces can include a Möbius strip, a Roman surface, a real projective plane, and a Klein bottle. In another example, the mathematical space can be an orientable space.

[0091] In some examples, step 542 may include obtaining a plurality of directions in the mathematical space. For example, the directions may be randomly selected. In other examples, the plurality of directions may be preselected directions, such as equally spaced directions. In yet other examples, the directions may be selected based on the distinctive visual content received by step 502. For example, a first direction may be selected when the distinctive visual content is a first visual content, and a second direction may be selected when the distinctive visual content is a second visual content. The second direction may be different from the first direction.

[0092] In some examples, step 544 may include obtaining a distinctive mathematical object in the mathematical space that corresponds to the distinctive visual content received by step 502. For example, a function of the distinctive visual content may be calculated to determine a distinctive mathematical object that corresponds to the distinctive visual content in the mathematical space. For example, the function may be a function of pixel values ​​of the distinctive visual content, may be a function of voxels of the distinctive visual content, may be a function of low-level features of the distinctive visual content, or may be a function of one or more characteristics of the distinctive visual content. Non-limiting examples of such functions may include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, and the like. In other examples, each mathematical object in the mathematical space may correspond to visual content. In still other examples, the distinctive mathematical object may be determined based on a mapping, for example, or utilizing a data structure that associates mathematical objects with visual content.

[0093] In some examples, step 546 may include, for each direction of the plurality of directions obtained by step 542, utilizing the unique mathematical object and the direction obtained by step 544 to determine a mathematical object in the mathematical space that corresponds to the unique mathematical object and the direction. For example, steps of a selected length in the direction may be obtained from the unique mathematical object to determine a mathematical object that corresponds to the unique mathematical object and the direction.

[0094] In some examples, step 548 may include generating visual content corresponding to a mathematical object in a mathematical space corresponding to the distinctive mathematical object and direction for each of the multiple directions obtained by step 542, thereby generating multiple variations of the distinctive visual content of step 522. For example, each mathematical object in the mathematical space may correspond to visual content by definition of the mathematical space. In other examples, a function of the mathematical object may be calculated to determine pixel values ​​for particular pixels of the visual content corresponding to the mathematical object or voxel values ​​for particular voxels of the visual content corresponding to the mathematical object.

[0095] 5D is a flow diagram of an example method 560 for identifying visual content utilized in training an inference model. In this example, method 560 may include computing a convolution of at least a portion of the distinctive visual content, thereby obtaining a resultant value of the computed convolution of at least a portion of the distinctive visual content (step 562), computing a mathematical function of the resultant value of the computed convolution of at least a portion of the distinctive visual content (step 564), selecting a threshold based on the data (step 566), comparing the threshold to the mathematical function of the resultant value of the computed convolution of at least a portion of the distinctive visual content (step 568), and determining a likelihood that the distinctive visual content is included in at least one training case of a plurality of training cases based on a result of the comparison of the threshold and the mathematical function of the resultant value of the computed convolution of at least a portion of the distinctive visual content (step 570).

[0096] In some examples, step 562 may include computing a convolution of at least a portion of the distinctive visual content received by step 502, thereby obtaining a resulting value of the computed convolution of at least a portion of the distinctive visual content. In some examples, step 564 may include computing a mathematical function of the resulting value (obtained by step 562) of the computed convolution of at least a portion of the distinctive visual content. Non-limiting examples of such functions may include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc. In some examples, step 566 may include selecting a threshold value based on the data accessed by step 504. For example, the threshold value may be a function of the data. Non-limiting examples of such functions may include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc. In other examples, step 566 may select a first threshold value when the data accessed by step 504 has a first value, and step 566 may select a second threshold value when the data accessed by step 504 has a second value. The second threshold value may be different from the first threshold value. In some examples, step 568 may include comparing the threshold selected by step 566 to the mathematical function (calculated by step 564) of the calculated convolution result value of at least a portion of the distinctive visual content obtained by step 562. In some examples, step 570 may include determining a likelihood of step 506 that the distinctive visual content is included in at least one training example of the plurality of training examples based on a result of the comparison (of step 568) of the threshold and the mathematical function of the calculated convolution result value of at least a portion of the distinctive visual content. For example, when the comparison results in a first result, step 570 may determine that the likelihood is a first likelihood, and when the comparison results in a second result, step 570 may determine that the likelihood is a second likelihood. The second likelihood may be different from the first likelihood.

[0097] Augmented reality environments can include non-photorealistic parts. Non-photorealistic parts reduce the immersive experience of the augmented reality environment. Therefore, it is useful to replace the non-photorealistic parts with photorealistic versions of the non-photorealistic parts. Manual identification of non-photorealistic parts as well as manual generation of such photorealistic versions and manual replacement of non-photorealistic parts with photorealistic versions is slow, costly and time-consuming, and completely impossible to perform in real time. Therefore, automatic identification of non-photorealistic parts, automatic generation of photorealistic versions, and automatic replacement of non-photorealistic parts with photorealistic versions are desirable.

[0098] Figures 6A, 6B, 6C, and 6D show example visual content. Figure 6A shows a schematic diagram of a bottle. Figure 6B shows a schematic diagram of a person. Figure 6C shows a photorealistic image of a bottle. Figure 6D shows a photorealistic image of a person. Figure 6E shows an example augmented reality environment. In this example, the augmented reality environment shows two people sitting around a table talking and drinking coffee, with a schematic visual representation of a bottle floating slightly above the table.

[0099] 7 is a flow diagram of an example method 700 of converting a non-photorealistic virtual environment into a photorealistic virtual environment. In this example, the method 700 can include receiving a first digital signal representing virtual content within an augmented reality environment (step 702), identifying a non-photorealistic portion of the virtual content using the first digital signal (step 704), analyzing the first digital signal using a generative model to generate a photorealistic version of the identified non-photorealistic portion of the virtual content (step 706), and generating a second digital signal configured to cause a wearable augmented reality device to present the generated photorealistic version in place of the identified non-photorealistic portion of the virtual content within the augmented reality environment (step 708). In one example, the entire augmented reality environment can be non-photorealistic and method 700 can include step 702, then step 706, then step 708, where step 706 can utilize a generative model to analyze the first digital signal to generate a photorealistic version of the entire augmented reality environment, and the second digital signal is configured to cause the wearable augmented reality device to present the photorealistic version of the entire augmented reality environment. In one example, the augmented reality environment can include other non-photorealistic portions not identified by step 704.

[0100] In one example, step 702 can receive a first digital signal representing virtual content in the augmented reality environment including a schematic representation of the bottle of FIG. 6A. Furthermore, step 704 can utilize the first digital signal to identify a non-photorealistic portion of the virtual content including the schematic representation of the bottle of FIG. 6A. Furthermore, step 706 can utilize the generative model to analyze the first digital signal to generate a photorealistic version of the bottle of FIG. 6C. Furthermore, step 708 can generate a second digital signal configured to cause the wearable augmented reality device to present the generated photorealistic version of the bottle of FIG. 6C in place of the non-photorealistic bottle in the augmented reality environment. In another example, step 702 can receive a first digital signal representing virtual content in the augmented reality environment including a schematic representation of the person of FIG. 6B. Furthermore, step 704 can utilize the first digital signal to identify a non-photorealistic portion of the virtual content including the schematic representation of the person of FIG. 6B. Furthermore, step 706 can utilize the generative model to analyze the first digital signal to generate a photorealistic version of the person of FIG. 6D. Further, step 708 can generate a second digital signal configured to cause the wearable augmented reality device to present a generated photo-realistic version of the person of FIG. 6D in place of the non-photo-realistic person in the augmented reality environment. In yet another example, step 702 can receive a first digital signal representing virtual content in the augmented reality environment of FIG. 6E including a photo-realistic visual representation of two people sitting around a table talking and drinking coffee, and a non-photo-realistic visual representation of a bottle floating slightly above the table. Further, step 704 can utilize the first digital signal to identify the non-photo-realistic visual representation of the bottle. Further, step 706 can utilize the generative model to analyze the first digital signal and generate a photo-realistic version of the bottle, e.g., generate a photo-realistic version of the bottle of FIG. 6C.Additionally, step 708 may generate a second digital signal configured to cause the wearable augmented reality device to present the generated photo-realistic version of the bottle of FIG. 6C in place of the non-photo-realistic bottle in the augmented reality environment and to reposition the bottle on the table as illustrated in FIG. 3B instead of the bottle floating slightly above the table.

[0101] In some examples, step 702 may include receiving a first digital signal representing virtual content within the augmented reality environment. For example, receiving the first digital signal may include reading the first digital signal from a memory, may include receiving the first digital signal from an external device (e.g., using a digital communication device), or may include generating the first digital signal (e.g., by analyzing color data and / or geometric data associated with the virtual content, e.g., using ray casting). In one example, the first digital signal may enable or include rendering of the virtual content. In one example, the first digital signal may include geometric data associated with the virtual content. In one example, the first digital signal may include three-dimensional data associated with the virtual content. In one example, the first digital signal may include color data associated with the virtual content. In one example, the first digital signal may include a representation of movement of the virtual content. In one example, the first digital signal can be or include a digital display signal configured to cause the wearable augmented reality device of step 708 to present virtual content within the augmented reality environment. In one example, the first digital signal can encode geometric data associated with the virtual content, such as a shape, polygon, plane, manifold, volume, etc.

[0102] In some examples, step 704 can utilize the first digital signal received by step 702 to identify non-photorealistic portions of the virtual content. In one example, the non-photorealistic portions of the virtual content can be, for example, a contiguous portion of the virtual content, can include multiple segmented fragments of the virtual content, can be or include a virtual object, can be or include a portion of a virtual object, can be or include a portion of a virtual background, can be the entire virtual content, can include some but not all of the virtual content, or can be the entire augmented reality environment. In one example, the identified non-photorealistic portions of the virtual content can be or include a virtual room (or a portion of a virtual room) within the augmented reality environment. In one example, the identified non-photorealistic portions of the virtual content can move within the augmented reality environment. In one example, the first digital signal can include indicating that a particular portion of the virtual content is non-photorealistic. In another example, step 704 can analyze the first digital signal to identify the non-photorealistic portions of the virtual content. For example, the machine learning model can be trained using training examples to identify non-photorealistic portions of the virtual content. Examples of such training examples can include sample virtual content along with labels indicating the non-photorealistic portions of the sample virtual content. Step 704 can analyze the first digital signal using the trained machine learning model to identify the non-photorealistic portions of the virtual content. In another example, step 704 can calculate a convolution of at least a portion of the virtual content, thereby obtaining a result value of the calculated convolution of at least a portion of the virtual content, and the result value of the calculated convolution of at least a portion of the virtual content can be used to identify the non-photorealistic portions of the virtual content. For example, when the result value is a first value, the non-photorealistic portions identified by step 704 can be a first portion of the virtual content, and when the result value is a second value, the non-photorealistic portions identified by step 704 can be a second portion of the virtual content. The second portion can be different from the first portion.In some examples, the first digital signal received by step 702 can indicate a field of view associated with the wearable augmented reality device of step 708, and step 704 can select a non-photorealistic portion of the virtual content based on the field of view associated with the wearable augmented reality device. For example, the non-photorealistic portion can be selected from within a field of view associated with the wearable augmented reality device, while non-photorealistic portions of the virtual content outside the field of view are excluded from the selected non-photorealistic portion. In other examples, the non-photorealistic portion can be selected from portions of the virtual content within the field of view or within a selected distance from the field of view, while non-photorealistic portions of the virtual content further away from the field of view are excluded from the selected non-photorealistic portion. In some examples, step 704 can utilize the first digital signal to cause the wearable augmented reality device of step 708 to present the virtual content to the individual, and can receive input from the individual to identify the non-photorealistic portion of the virtual content. For example, the input can be received via a user interface, via gestures, via voice commands, or via an input device associated with the wearable augmented reality device.

[0103] In some examples, step 706 can utilize a generative model to analyze the first digital signal received by step 702 to generate a photo-realistic version of the non-photo-realistic portion of the virtual content identified by step 704. For example, the generative model can be a generative model trained using training examples to generate a photo-realistic version of the non-photo-realistic portion of the virtual content. Examples of such training examples can include sample virtual content including the sample non-photo-realistic portion along with a photo-realistic version of the sample non-photo-realistic portion. Step 706 can utilize the trained generative model to analyze the first digital signal received by step 702 to generate a photo-realistic version of the non-photo-realistic portion of the virtual content identified by step 704. In some examples, step 706 can analyze the first digital signal to determine a mathematical object in a mathematical space that corresponds to the non-photo-realistic portion of the identified virtual content. In one example, the mathematical space can be a non-orientable space. Non-limiting examples of such non-orientable spaces can include a Moebius strip, a Roman surface, a real projective plane, and a Klein bottle. In other examples, the mathematical space can be an orientable space. In one example, a function of the non-photorealistic portion of the virtual content can be calculated to determine a mathematical object corresponding to the identified non-photorealistic portion of the virtual content. For example, the function can be a function of pixel values ​​of the non-photorealistic portion of the virtual content, a function of voxels of the non-photorealistic portion of the virtual content, a function of low-level features of the non-photorealistic portion of the virtual content, a function of a geometric shape of the non-photorealistic portion of the virtual content, a function of one or more characteristics of the non-photorealistic portion of the virtual content, or a function of a portion of the first digital signal that corresponds to the non-photorealistic portion of the virtual content. Non-limiting examples of such functions can include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc. In another example, each mathematical object in the mathematical space can correspond to a virtual content, and the determined mathematical object can be the mathematical object that corresponds to the non-realistic portion.The mathematical object may be determined, for example, based on a mapping or using a data structure that associates mathematical objects with virtual content. In some examples, step 706 may utilize a mathematical object in a mathematical space that corresponds to the identified non-photorealistic portion of the virtual content to generate a photorealistic version of the identified non-photorealistic portion of the virtual content. For example, the photorealistic version may be a function of the mathematical object. Non-limiting examples of such functions may include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc. In other examples, a function of the mathematical object that corresponds to the identified non-photorealistic portion may be calculated to obtain a second mathematical object (e.g., in the same mathematical space or in another mathematical space), and step 706 may utilize the second mathematical object to generate a photorealistic version of the identified non-photorealistic portion of the virtual content. For example, the photorealistic version may be a function of the second mathematical object. In some examples, step 706 can determine a category of the non-photorealistic portion of the virtual content by analyzing the first digital signal, e.g., by utilizing a classification model to classify the non-photorealistic portion into one of a plurality of alternative classes, and each class is associated with a category. Additionally, step 706 can select a generative model of a plurality of generative models using the category of the non-photorealistic portion of the virtual content. For example, step 706 can use the category to access a data structure that associates categories with generative models to select the generative model.

[0104] In some examples, step 708 may include generating a second digital signal configured to cause the wearable augmented reality device to present the photo-realistic version generated by step 706 in place of the non-photo-realistic portion of the virtual content identified by step 704 in the augmented reality environment of step 702. In one example, the second digital signal may enable or include a rendering of the photo-realistic version generated by step 706. In one example, the second digital signal may include geometric data associated with the photo-realistic version generated by step 706. In one example, the second digital signal may include three-dimensional data associated with the photo-realistic version generated by step 706. In one example, the second digital signal may include color data associated with the photo-realistic version generated by step 706. In one example, the second digital signal may include a representation of a movement of the photo-realistic version generated by step 706. In one example, the second digital signal can include indicating a non-photorealistic portion of the virtual content identified by step 704, e.g., to enable presentation of a photorealistic version generated by step 706 in place of the non-photorealistic portion of the virtual content identified by step 704 in the augmented reality environment of step 702. In one example, the second digital signal can include indicating one or more locations within the augmented reality environment of step 702 for presentation of the photorealistic version generated by step 706. In one example, step 708 can utilize the second digital signal to store information encoded within the second digital signal in a memory. In one example, step 708 can transmit the second digital signal to an external device (e.g., a wearable augmented reality device, a computer-controlled device that adjusts the augmented reality environment, etc.). In some examples, the virtual content of step 702 can include a second non-photorealistic portion that is different from the non-photorealistic portion of the virtual content identified by step 704.Further, the second digital signal generated by step 708 is configured to cause the wearable augmented reality device to present the photo-realistic version generated by step 706 in place of the non-photo-realistic portion of the virtual content identified by step 704 while still presenting the second non-photo-realistic portion. In some examples, step 708 can switch between causing the wearable augmented reality device to present the generated photo-realistic version in place of the identified non-photo-realistic portion of the virtual content in the augmented reality environment based on input from the individual utilizing the wearable augmented reality device, and causing the wearable augmented reality device to present the virtual content including the identified non-photo-realistic portion of the virtual content. For example, the input is received via a user interface, via gestures, via voice commands, or via an input device associated with the wearable augmented reality device.

[0105] In some examples, the non-photorealistic portion of the virtual content identified by step 704 may include an avatar associated with an individual, the photorealistic version (generated by step 706) of the identified non-photorealistic portion of the virtual content is a photorealistic depiction of the individual, and the analysis using the generative model (by step 706) is based on a photograph of the individual. For example, the generative model may be a generative model trained using training examples to generate a photorealistic version of the non-photorealistic avatar based on an image of the individual. Examples of such training examples may include sample images of a sample individual and sample virtual content including the sample non-photorealistic avatar, along with a photorealistic version of the sample non-photorealistic avatar, where the photorealistic version visually appears as the sample individual. Step 706 may analyze the first digital signal received by step 702 using the trained generative model to generate a photorealistic version of the non-photorealistic avatar based on a photograph of the individual. In some examples, the non-photorealistic portion of the virtual content identified by step 704 can include a non-photorealistic skin color, and step 706 can replace the non-photorealistic skin color with a photorealistic skin color, thereby generating a photorealistic version of the identified non-photorealistic portion of the virtual content replacing the non-photorealistic skin color with the photorealistic skin color. In some examples, the non-photorealistic portion of the virtual content identified by step 704 can include a first element, a second element, and a non-photorealistic ratio between a dimension of the first element and a dimension of the second element. Further, generating the photorealistic version by step 706 can include resizing at least one of the first element or the second element. In one example, the photorealistic version of the identified non-photorealistic portion of the virtual content can include a photorealistic ratio between a dimension of the first element and a dimension of the second element as a result of the resizing. In some examples, the non-photorealistic portion of the virtual content identified by step 704 can include a non-photorealistic landscape.Further, generating a photo-realistic version by step 706 may include replacing a non-photo-realistic landscape with a photo-realistic landscape, thereby generating a photo-realistic version of the non-photo-realistic portion of the identified virtual content replacing the non-photo-realistic landscape with the photo-realistic landscape.

[0106] More and more visual content is being generated artificially, for example, using generative models. However, while some of the generated visual content is photorealistic, others are non-photorealistic, e.g., due to limitations of the generative model, even if the original intention was to generate a photorealistic version. It is useful to identify which of the artificially generated visual content is photorealistic and which is non-photorealistic, e.g., to avoid using non-photorealistic visual content or to evaluate and improve generative models. Computer-controlled models may fail to distinguish some of the non-photorealistic visual content from the photorealistic visual content. Therefore, it is useful to involve a human in the process. Automatically managing a manual process improves efficiency and reduces errors.

[0107] 8 is a diagram illustrating an example user interface 800. In this example, user interface 800 includes a presentation of four visual contents 802, 804, 806, 808, along with a control disposed next to each visual content that allows a user to mark the visual content as real or artificial. Additionally or alternatively, the control can allow a user to indicate certainty of marking the visual content as real or artificial.

[0108] FIG. 9A is a flow diagram of an example method 900 for determining realism of artificially generated visual content. In this example, the method 900 may include accessing one or more artificially generated visual content, where the one or more artificially generated visual content includes a particular artificially generated visual content (step 902); accessing one or more captured visual content, where each visual content of the one or more captured visual content is captured from a real physical environment using an image sensor (step 904); for each person of the plurality of people, presenting a mix of the plurality of visual contents to the person, where the mix of the plurality of visual contents includes at least one visual content of the one or more artificially generated visual contents and at least one visual content of the one or more captured visual contents, where the mix of the plurality of visual contents may include the particular artificially generated visual content (step 906); for each person of the plurality of people, receiving from the person a response of the presentation to the person, where for each visual content of the set of at least one mix of visual content, the response may indicate whether the person believed the visual content was artificially generated visual content (step 908); and determining a realism of the particular artificially generated visual content based on the response (step 910).

[0109] In some examples, step 902 may include accessing one or more artificially generated visual contents. The one or more artificially generated visual contents may include a particular artificially generated visual content. For example, accessing one or more artificially generated visual contents by step 902 may include accessing one or more artificially generated visual contents in a digital memory, may include communicating with an external device that holds the one or more artificially generated visual contents (e.g., using a digital communication device), or may include generating one or more artificially generated visual contents (e.g., using a generative model). In one example, the artificially generated visual content of each of the one or more artificially generated visual contents is generated using a conditional generative adversarial network, e.g., the artificially generated visual content of each of the one or more artificially generated visual contents is generated using a conditional generative adversarial network having different input conditions. In some examples, for each visual content of the one or more artificially generated visual contents, step 902 may analyze the visual content using a machine learning model to determine a level of certainty that the visual content is photorealistic. For example, the one or more artificially generated visual contents may include only visual content that an automated analysis (e.g., utilizing the same or a different machine learning model) has determined to be photo-realistic. In one example, the machine learning model may be a machine learning model trained using training examples to determine a level of confidence that the visual content is photo-realistic. Examples of such training examples may include sample visual content along with labels indicating a level of confidence that the sample visual content is photo-realistic. Step 902 may analyze the visual content using the trained machine learning model to determine a level of confidence that the visual content is photo-realistic.Further, step 902 may select the particular artificially-generated visual content from the one or more artificially-generated visual contents based on a level of certainty. For example, step 902 may select the visual content associated with the lowest certainty from the one or more artificially-generated visual contents as the particular artificially-generated visual content. In another example, step 902 may select the visual content associated with the lowest certainty from the one or more artificially-generated visual contents determined to be realistic by automated analysis (e.g., using the same or a different machine learning model) as the particular artificially-generated visual content.

[0110] In some examples, step 904 can include accessing one or more captured visual contents, where each visual content of the one or more captured visual contents is captured from a real physical environment using an image sensor. For example, accessing the one or more captured visual contents, per step 904, can include accessing the one or more captured visual contents in a digital memory, can include communicating with an external device that holds the one or more captured visual contents (e.g., using a digital communication device), or can include capturing the one or more captured visual contents using at least one image sensor.

[0111] In some examples, step 906 may include, for each person of the plurality of people, presenting a mix of the plurality of visual contents to the person. The mix of the plurality of visual contents may include at least one visual content of the one or more artificially generated visual contents accessed by step 902 and at least one visual content of the one or more captured visual contents accessed by step 904. The mix of the plurality of visual contents may include a particular artificially generated visual content accessed by step 902. For example, the mix of the plurality of visual contents is presented to the person via a user interface (e.g., via user interface 800 or via a different user interface). In one example, the mix of the plurality of visual contents presented by step 906 to one of the plurality of people is identical to the mix of the plurality of visual contents presented by step 906 to other people of the plurality of people. In other examples, the mix of the plurality of visual contents presented by step 906 to one of the plurality of people may be different from the mix of the plurality of visual contents presented by step 906 to other people of the plurality of people.

[0112] In some examples, step 908 may include, for each person of the plurality of people, receiving a response from the person to the presentation of step 906. For each visual content of the mix of at least one visual content of the group of step 906, the response may indicate whether the person believed the visual content was artificially generated visual content. The mix of at least one visual content of the group of step 906 may include, for example, one visual content, may include one or more visual content, may include two visual content, may include two or more visual content, may include three or more visual content, may include ten or more visual content, or may include a particular artificially generated visual content accessed by step 902. Indicating whether the person believed the visual content was realistic is understood to indicate whether the person believed the visual content was artificially generated visual content. In one example, the response may be received via a user interface (e.g., via a user interface utilized by step 906 to present the mix to the person, via user interface 800, or via a different user interface). In other examples, the response can be received, for example, via voice commands, via gestures, via text input, via a pointing device, via a keyboard, via an image sensor, via an audio sensor, etc. In one example, the response can further indicate a certainty that the person believed the visual content to be artificially generated visual content.

[0113] In some examples, step 910 can include determining the realism of the particular artificially-generated visual content accessed by step 902 based on the responses received by step 908. For example, step 910 can calculate a function of the responses to determine the realism of the particular artificially-generated visual content. Non-limiting examples of such functions can include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc. In one example, the function can be invariant across people, such as an average of the responses. In another example, the function can treat different people's responses differently. For example, each person is associated with a weight, and the function can be a weighted average of the responses. In one example, a machine learning model can be trained using training examples to determine the realism from the responses. Examples of such training examples can include sample responses associated with the sample visual content along with a label indicating the realism of the sample visual content. Step 910 can utilize the trained machine learning model to analyze the responses received by step 908 to determine the realism of the particular artificially-generated visual content accessed by step 902. In one example, the realism of a particular artificially generated visual content determined by step 910 can be an absolute degree. Non-limiting examples of such absolute degrees can include "very realistic", "realistic", "not realistic", "absurd", a score within a predefined range (8.2 in a range of 0-10), a percentage (71% of people believed it was realistic), etc. In another example, the realism of a particular artificially generated visual content determined by step 910 can be a relative degree to at least one other artificially generated visual content. Non-limiting examples of such relative degrees can include "more realistic", "less realistic", "as realistic", "most realistic of the visual content set", "least realistic of the visual content set", "30% more realistic", "20% less realistic", etc.In one example, the realism of the particular artificially generated visual content determined by step 910 can be based on a comparison of the response to the particular artificially generated visual content and at least one response to other artificially generated visual content of the one or more artificially generated visual content accessed by step 902. For example, the comparison can indicate that more people found the particular artificially generated visual content to be more realistic than other artificially generated visual content, and accordingly, the realism can be "more realistic" or even improved. In another example, the realism of the particular artificially generated visual content determined by step 910 can be based on a comparison of the response to the particular artificially generated visual content and at least one response to visual content of the one or more captured visual content accessed by step 904. For example, the comparison can indicate that fewer people found the particular artificially generated visual content to be more realistic than visual content of the one or more captured visual content, and accordingly, the realism can be "less realistic" or even improved.

[0114] In some examples, method 900 may further include determining whether to remove the particular artificially generated visual content accessed by step 902 based on the realism of the particular artificially generated visual content determined by step 910. For example, method 900 may remove the particular artificially generated visual content when the determined realism is below a selected value, and method 900 may avoid removing the particular artificially generated visual content when the determined realism is above a selected value. In one example, the selected value may be pre-selected (e.g., by a user or by configuration). In other examples, the selected value may be based on the particular artificially generated visual content. In yet other examples, the selected value may be based on the determined realism of at least one other visual content. In one example, deleting the particular artificially-generated visual content determined by step 910 may include at least one of deleting the particular artificially-generated visual content from memory, removing the particular artificially-generated visual content from a data structure (or from a database), causing an external device to delete the particular artificially-generated visual content (e.g., by sending a digital signal to the external device), or marking in memory that the particular artificially-generated visual content has been deleted.

[0115] In some examples, the method 900 may further include determining whether to utilize the particular artificially generated visual content accessed by step 902 for a particular application based on the determined realism of the particular artificially generated visual content determined by step 910. For example, when the determined realism is above a selected value, the method 900 may utilize the particular artificially generated visual content for a particular application, and when the determined realism is below the selected value, the method 900 may avoid utilizing the particular artificially generated visual content for a particular application. In one example, the selected value may be pre-selected (e.g., by a user or by configuration). In another example, the selected value may be based on the particular artificially generated visual content. In yet another example, the selected value may be based on the determined realism of at least one other visual content. Non-limiting examples of such particular applications may include, for example, storing the visual content in a memory, presenting the visual content to an individual (e.g., via a user interface or via a physical display screen), or analyzing the visual content to determine information.

[0116] In some examples, the method 900 may further include analyzing the particular artificially generated visual content accessed by step 902 to determine at least one reason for the realism of the particular artificially generated visual content determined by step 910. In one example, the analysis may be based on the response received by step 908. In one example, the analysis may be based on the realism of the particular artificially generated visual content determined by step 910. In one example, a machine learning model may be trained using training examples to determine reasons for the realism of the visual content. Examples of such training examples may include sample visual content associated with the sample realism along with labels indicating one or more reasons for the sample realism. The trained machine learning model may be used to analyze the particular artificially generated visual content accessed by step 902 to determine at least one reason for the realism of the particular artificially generated visual content determined by step 910. In one example, a convolution of at least a portion of the particular artificially generated visual content accessed by step 902 may be calculated, thereby obtaining a result value of the calculated convolution of at least a portion of the particular artificially generated visual content. Further, the result value of the calculated convolution of at least a portion of the particular artificially generated visual content may be utilized to determine at least one reason for the determined realism of the particular artificially generated visual content. For example, when the result value is a first value, a first reason for the determined realism is determined, and when the result value is a second value, a second reason for the determined realism is determined. The second reason may be different from the first reason. In one example, the method 900 may further include presenting the determined at least one reason to the individual, for example, visually, audibly, graphically, or textually via a user interface.In one example, the method 900 may further include generating new artificially generated visual content utilizing the at least one reason. For example, the new artificially generated visual content is generated to overcome the at least one reason. In one example, when the reason is "top left corner is too dark", the particular artificially generated visual content accessed by step 902 is transformed to make the top left corner brighter. In another example, when the reason is "face skin color does not match body skin color", the face skin color in the particular artificially generated visual content accessed by step 902 is modified to match the body skin color. In yet another example, when the reason is "proportion between arm size and body size is wrong", at least one of the arms and the body in the particular artificially generated visual content accessed by step 902 is resized and proportions are corrected.

[0117] 9B is a flow diagram of an example method 920 for determining realism of artificially generated visual content. In some examples, the one or more artificially generated visual contents accessed by step 902 can include a unique artificially generated visual content, where the unique artificially generated visual content can be different from the particular artificially generated visual content accessed by step 902. The method 920 can include determining the realism of the unique artificially generated visual content based on at least one response to the unique artificially generated visual content (step 922), comparing data associated with the unique artificially generated visual content with data associated with the particular artificially generated visual content (step 924), and determining at least one reason for the determined realism of the particular artificially generated visual content based on a result of the comparison (step 926).

[0118] In some examples, step 922 may include determining the realism of the unique artificially generated visual content included in the one or more artificially generated visual contents accessed by step 902 based on at least one response to the unique artificially generated visual content, e.g., as described above for method 900 and the particular artificially generated visual content accessed by step 902. The unique artificially generated visual content may be different from the particular artificially generated visual content accessed by step 902. In one example, the unique artificially generated visual content may be included in all mixes presented by step 906, and the at least one response to the unique artificially generated visual content may be a response received by step 908. In another example, the unique artificially generated visual content may be included in a particular mix presented to a particular person by step 906, and the at least one response to the unique artificially generated visual content may include a response received from the particular person by step 908. In yet another example, the at least one response to the unique artificially generated visual content may include one or more responses not included in the responses received by step 908. In an additional example, the responses received by step 908 can include one or more responses not included in the at least one response to the unique artificially-generated visual content.

[0119] In some examples, step 924 can include comparing data associated with the unique artificially-generated visual content of step 922 with data associated with the particular artificially-generated visual content accessed by step 902. For example, step 924 can include comparing pixel data associated with the unique artificially-generated visual content of step 922 with pixel data associated with the particular artificially-generated visual content accessed by step 902. In other examples, step 924 can analyze the visual content of each of the unique and particular artificially-generated visual content utilizing a visual object detection algorithm to identify objects depicted in the visual content, and can compare the objects depicted in the unique artificially-generated visual content of step 922 with the objects depicted in the particular artificially-generated visual content accessed by step 902. For example, comparing the objects can include comparing at least one of the object type, quantity, size, appearance (such as brightness or color), location, or spatial orientation. In yet another example, step 924 can utilize a visual event detection algorithm to analyze the visual content of each of the unique and particular artificially-generated visual content to identify events depicted in the visual content, and can compare the events depicted in the unique artificially-generated visual content of step 922 with the events depicted in the particular artificially-generated visual content accessed by step 902. For example, comparing the events can include comparing at least one of the type, duration, timing, location, or objects associated with the events.

[0120] In some examples, each of the one or more artificially generated visual contents accessed by step 902 can be a result of applying a function to different input values. One non-limiting example of such a function can include a conditional generative adversarial network, and the input values ​​can be input conditions. Furthermore, the data associated with the particular artificially generated visual content of step 902 can be or include input values ​​associated with the particular artificially generated visual content, and the data associated with the particular artificially generated visual content of step 922 can be or include input values ​​associated with the particular artificially generated visual content. In this example, step 924 can compare the input values ​​associated with the particular artificially generated visual content of step 922 to the input values ​​associated with the particular artificially generated visual content accessed by step 902. In some examples, step 924 can analyze the particular artificially generated visual content accessed by step 902 to determine the data associated with the particular artificially generated visual content, and / or can analyze the particular artificially generated visual content of step 922 to determine the data associated with the particular artificially generated visual content. For example, a machine learning model can be trained using training examples to determine the data based on the visual content. Such training example cases may include sample visual content along with labels indicative of data associated with the visual content. Step 924 may utilize the trained machine learning model to analyze particular artificially-generated visual content to determine data associated with the particular artificially-generated visual content and / or to analyze unique artificially-generated visual content to determine data associated with the particular artificially-generated visual content.In some examples, step 924 may compute a convolution of at least a portion of the particular artificially-generated visual content accessed by step 902, thereby determining data related to the particular artificially-generated visual content, and / or may compute a convolution of at least a portion of the particular artificially-generated visual content of step 922, thereby determining data related to the particular artificially-generated visual content. For example, the data related to the visual content may be or include a resultant value of the convolution of at least a portion of the visual content, or may be a function of or include a resultant value of the convolution of at least a portion of the visual content.

[0121] In some examples, step 926 may include determining at least one reason for the determined realism of the particular artificially generated visual content based on the result of the comparison of step 924. In some examples, step 926 may determine at least one reason for the determined realism of the particular artificially generated visual content based on the result of the comparison of step 924, the realism of the particular artificially generated visual content determined by step 910, and the realism of the particular artificially generated visual content determined by step 922. In some examples, the comparison of step 924 may identify a visual difference between the particular artificially generated visual content of step 922 and the particular artificially generated visual content accessed by step 902, and the difference in realism between the two may be attributed to a visual difference. For example, the visual difference may be a difference in skin color of a person depicting both the particular and the particular artificially generated visual content, and the reason may be "unrealistic skin color." In another example, the visual difference can be a difference in the size of the limbs of a person depicting both the specific and particular artificially-generated visual content, and the reason can be "unrealistic limb sizes."

[0122] In one example, method 920 can further include presenting the determined at least one reason to the individual, e.g., visually, audibly, graphically, or textually, via a user interface. In one example, method 920 can further include generating new artificially generated visual content utilizing the at least one reason, e.g., as described above for method 900.

[0123] 9C is a flow diagram of an example method 940 for determining realism of artificially-generated visual content. In some examples, the one or more artificially-generated visual contents accessed by step 902 can include unique artificially-generated visual content, and the unique artificially-generated visual content can be different from the particular artificially-generated visual content accessed by step 902. The method 940 may include determining a realism of the particular artificially generated visual content based on at least one response to the particular artificially generated visual content (step 922); determining a particular mathematical object in the mathematical space associated with the particular artificially generated visual content (step 944); determining a particular mathematical object in the mathematical space associated with the particular artificially generated visual content (step 946); determining a new mathematical object in the mathematical space using the determined realism of the particular artificially generated visual content, the determined realism of the particular artificially generated visual content, the particular mathematical object, and the particular mathematical object, where the new mathematical object may be different from the particular mathematical object and the particular mathematical object (step 948); and generating new artificially generated visual content using the new mathematical object in the mathematical space (step 950).

[0124] In some examples, a mathematical object in a mathematical space associated with the artificially generated visual content can be determined. For example, step 944 can include determining a particular mathematical object in a mathematical space associated with the particular artificially generated visual content accessed by step 902. In other examples, step 946 can include determining a particular mathematical object in a mathematical space associated with the particular artificially generated visual content of step 922. In one example, the mathematical space can be a non-orientable space. Non-limiting examples of such non-orientable spaces can include a Möbius strip, a Roman surface, a real projective plane, and a Klein bottle. In other examples, the mathematical space can be an orientable space. In some examples, the artificially generated visual content can be analyzed to determine a mathematical object in a mathematical space associated with the artificially generated visual content. For example, a function of the artificially generated visual content can be calculated to thereby determine a mathematical object associated with the artificially generated visual content. For example, the function can be a function of pixel values ​​of the artificially generated visual content, a function of voxels of the artificially generated visual content, a function of low-level features of the artificially generated visual content, or a function of one or more characteristics of the artificially generated visual content. Non-limiting examples of such functions can include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc. In other examples, the mathematical object is determined based on, for example, a mapping or utilizing a data structure that associates the mathematical object with the visual content.

[0125] In some examples, step 948 may include determining a new mathematical object in the mathematical space using the realism of the particular artificially-generated visual content determined by step 910, the realism of the particular artificially-generated visual content determined by step 922, the particular mathematical object determined by step 944, and the particular mathematical object determined by step 946. The new mathematical object may be different from the particular mathematical object and the particular mathematical object. For example, the new mathematical object may be a weighted average of the particular mathematical object determined by step 944 and the particular mathematical object determined by step 946, where the weights are selected based on the realism of the particular artificially-generated visual content determined by step 910 and the realism of the particular artificially-generated visual content determined by step 922. In other examples, the new mathematical object can be a function of the realism of the particular artificially-generated visual content determined by step 910, the realism of the particular artificially-generated visual content determined by step 922, the particular mathematical object determined by step 944, and the particular mathematical object determined by step 946. Non-limiting examples of such functions can include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc.

[0126] In some examples, step 950 may include generating new artificially generated visual content utilizing the new mathematical objects in the mathematical space determined by step 948. For example, the new artificially generated visual content (or at least one pixel value of the artificially generated visual content) may be a function of the new mathematical objects. Non-limiting examples of such functions may include linear functions, non-linear functions, exponential functions, polynomial functions, logarithmic functions, etc. In other examples, each mathematical object in the mathematical space may correspond to visual content, and the new artificially generated visual content may be visual content corresponding to the new mathematical objects. In other examples, step 950 may utilize the new mathematical objects to access a data structure relating mathematical objects to visual content to determine the new artificially generated visual content.

[0127] The above description is provided for illustrative purposes, and is not exhaustive or limited to the precise forms or embodiments disclosed. Various modifications and adaptations will be apparent to those skilled in the art in light of the practice and specifications of the disclosed embodiments. Additionally, although features of the disclosed embodiments are described as being stored in memory, those skilled in the art will appreciate that these features can also be stored in other types of computer readable media, such as secondary storage devices, such as hard disks or CD-ROMs, or other forms of RAM or ROM, USB media, DVDs, Blu-ray, Ultra HD Blu-ray, or other optical drive media.

[0128] Computer programs based on the above description and disclosed methods are within the skill of an experienced developer. Various programs or program modules can be created or designed in conjunction with existing software using any technology known to those skilled in the art. For example, program parts or program modules can be designed in or with the .Net Framework, .Net Compact Framework (and related languages ​​such as Visual Basic, C, etc.), Java, C++, Objective-C, HTML, HTML / AJAX combination, XML, or HTML including Java applets.

[0129] Moreover, while exemplary embodiments are described herein, the scope of any and all embodiments will be understood by one of ordinary skill in the art based on this disclosure to include equivalent elements, modifications, omissions, combinations (e.g., of features of the various embodiments), adaptations, and / or alterations. The examples are understood to be non-exclusive examples. Furthermore, the steps of the disclosed methods can be modified in any manner, for example, by changing the order of steps and / or by inserting or deleting steps. Accordingly, the specification and examples are considered to be illustrative only. [Explanation of symbols]

[0130] 132 Communication Hardware 300 Multiple Visual Content 302 Visual content of two people sitting around a table talking and drinking coffee 304 Visual Content of the Bottle 306 Visual content of standing people 308 Car Visual Content 310 different car visual content 312 Bus Visual Content 314 Tractor Visual Content 350 Exemplary Visual Content 400 Exemplary Methods for Attributing Generated Visual Content to Training Examples 420 Exemplary Method for Attributing Generated Visual Content to Training Examples and Selecting Visual Content for Particular Uses Based on Attribution 440 Exemplary Method for Attributing Generated Visual Content to Training Examples and Selecting Visual Content for Particular Uses Based on Attribution 460 Exemplary Method for Attributing Generated Visual Content to Training Examples 480 Exemplary Method for Attributing Generated Visual Content to Training Examples 500 Exemplary Method for Identifying Visual Content Used to Train an Inference Model 520 Exemplary Method for Identifying Visual Content Used to Train Inference Models 540 An Exemplary Method for Identifying Visual Content Used to Train an Inference Model 560 An Exemplary Method for Identifying Visual Content Used to Train an Inference Model 700 Exemplary Method for Converting a Non-Photorealistic Virtual Environment to a Photorealistic Virtual Environment 800 Exemplary User Interface 802 Visual Content 804 Visual Content 806 Visual Content 808 Visual Content 900 Illustrative Method for Determining Realism of Artificially Generated Visual Content 920 Illustrative Method for Determining Realism of Artificially Generated Visual Content 940 Illustrative Method for Determining Realism of Artificially Generated Visual Content

Claims

1. 1. A non-transitory computer-readable medium comprising instructions that cause at least one processor to perform operations for attributing generated visual content to training examples, the operations comprising: receiving first visual content generated using a generative model, the generative model being a result of training a machine learning algorithm using a plurality of training cases, each training case of the plurality of training cases being associated with the visual content; determining one or more characteristics of the first visual content; For each training case of the plurality of training cases, analyzing the visual content associated with the training case to determine one or more characteristics of the visual content associated with the training case; utilizing one or more characteristics of the first visual content and characteristics of the visual content associated with the plurality of training examples to attribute the first visual content to a first subgroup of at least one, but not all, of the plurality of training examples; determining whether at least one of the visual contents associated with the training examples of the first subgroup is associated with at least one first resource; and for each of said at least one first resource, updating a data record associated with said resource based on the attribution; A non-transitory computer-readable medium comprising:

2. The operation further comprises: determining, for each training example of the first subgroup, a degree of membership of the first visual content to the training examples using one or more characteristics of the first visual content and characteristics of the visual content associated with the first subgroup; and and for each resource of the at least one first resource, further updating the data record associated with the resource based on a degree of membership associated with the resource.

10. The non-transitory computer-readable medium of claim 1, further comprising:

3. 2. The non-transitory computer-readable medium of claim 1 , wherein determining one or more characteristics of the first visual content is based on intermediate results of the generative model during generation of the first visual content.

4. The operation further comprises: calculating a convolution of at least a portion of the first visual content, thereby obtaining a calculated convolution result value of at least a portion of the first visual content; and determining one or more characteristics of the first visual content based on the resulting value of the computed convolution of at least a portion of the first visual content; 10. The non-transitory computer-readable medium of claim 1, further comprising:

5. The operation further comprises: analyzing the first visual content to detect at least a first object and a second object depicted in the first visual content; and determining one or more characteristics of the first visual content based on a position of the first object within the first visual content and a position of the second object within the first visual content.

10. The non-transitory computer-readable medium of claim 1, further comprising:

6. The operation further comprises: Analyzing the first visual content to determine a type of object depicted in the first visual content; and determining one or more characteristics of the first visual content based on the type of the object; 10. The non-transitory computer-readable medium of claim 1, further comprising:

7. The operation further comprises: Analyzing the first visual content to detect an event depicted in the first visual content; and determining one or more characteristics of the first visual content based on the event; 10. The non-transitory computer-readable medium of claim 1, further comprising:

8. 2. The non-transitory computer-readable medium of claim 1 , wherein one or more characteristics of the first visual content are based on a temporal relationship between an appearance of a first object and an appearance of a second object in the first visual content.

9. 2. The non-transitory computer-readable medium of claim 1, wherein the first visual content includes at least a first frame and a second frame, and one or more characteristics of the first visual content are based on a change in a spatial relationship between two objects that appear in the first visual content from the first frame to the second frame.

10. 2. The non-transitory computer-readable medium of claim 1 , wherein training the machine learning algorithm to generate the generative model includes an iterative process, in which in each iteration of the iterative process, a training example of the plurality of training examples is analyzed to update a loss function, and one or more characteristics of the visual content associated with a particular training example are based on updating the loss function in an iteration that includes analysis of the particular training example.

11. Training the machine learning algorithm to generate the generative model includes a first step of training using a first subgroup of the plurality of training examples to obtain an intermediate model, and a second step of training using a second subgroup of the plurality of training examples and the intermediate model for initialization to obtain the generative model, the second subgroup being different from the first subgroup, and the operation further comprises: comparing results associated with the first visual content and the intermediate model to results associated with the first visual content and the generative model; and for each training example of the second subgroup, determining whether to attribute the first visual content to the training example based on a result of the comparison; 10. The non-transitory computer-readable medium of claim 1, further comprising:

12. 2. The non-transitory computer-readable medium of claim 1 , wherein one or more characteristics of particular visual content associated with a particular training example of the plurality of training examples are determined based on a convolution of at least a portion of the particular visual content.

13. 2. The non-transitory computer-readable medium of claim 1 , wherein one or more characteristics of particular visual content associated with a particular training example of the plurality of training examples are determined based on a temporal relationship between an appearance of a first object and an appearance of a second object within the particular visual content.

14. 2. The non-transitory computer-readable medium of claim 1 , wherein one or more characteristics of a particular visual content associated with a particular training example of the plurality of training examples is determined based on a change in a spatial relationship of two objects that appear within the particular visual content from the first frame of the particular visual content to the second frame of the particular visual content.

15. The operation further comprises: receiving second visual content generated utilizing the generative model; determining one or more characteristics of the second visual content; utilizing one or more characteristics of the second visual content and characteristics of the visual content associated with the plurality of training examples to attribute the second visual content to a second subgroup of at least one, but not all, of the plurality of training examples, the second subgroup including at least one training example not included in the first subgroup; determining whether at least one of the visual contents associated with the training examples of the second subgroup is associated with at least one second resource, the at least one second resource including one or more resources not included in the at least one first resource; refraining from utilizing the second visual content based on the at least one second resource; and Initiating utilization of the first visual content; 10. The non-transitory computer-readable medium of claim 1, further comprising:

16. The operation further comprises: receiving second visual content generated utilizing the generative model; determining one or more characteristics of the second visual content; utilizing one or more characteristics of the second visual content and characteristics of the visual content associated with the plurality of training examples to attribute the second visual content to a second subgroup of at least one, but not all, of the plurality of training examples, the second subgroup including at least one training example not included in the first subgroup; Access a data structure that associates visual content with quantity; utilizing the data structure to determine whether at least one of the visual contents associated with the training examples of the first subgroup is associated with a first quantity; utilizing the data structure to determine whether at least one of the visual contents associated with the training examples of the second subgroup is associated with a second quantity; refraining from using the second visual content based on the first amount and the second amount and commencing using the first visual content; and updating a data record associated with the resource based on the first amount; 10. The non-transitory computer-readable medium of claim 1, further comprising:

17. The operation further comprises: Embedding the first visual content within a mathematical space utilizing one or more characteristics of the first visual content; For each training example of the plurality of training examples, embedding the visual content associated with the training example within the mathematical space using one or more characteristics of the visual content associated with the training example; and utilizing the mathematical space to select the first subgroup of at least one, but not all, of the plurality of training examples.

10. The non-transitory computer-readable medium of claim 1, further comprising:

18. The operation further comprises: utilizing the parameters of the generative model and characteristics of the visual content associated with the plurality of training examples to impute the parameters of the generative model to a second subgroup of at least one, but not all, of the plurality of training examples; determining whether at least one of the visual contents associated with the training examples of the second subgroup is associated with at least one second resource; and for each of the at least one second resource, updating a data record associated with the resource based on the attribution of the parameters of the Generative Model; 10. The non-transitory computer-readable medium of claim 1, further comprising:

19. 1. A system for attributing generated visual content to training examples, the system comprising at least one processor, the at least one processor comprising: receiving first visual content generated using a generative model, the generative model being a result of training a machine learning algorithm using a plurality of training cases, each training case of the plurality of training cases being associated with the visual content; determining one or more characteristics of the first visual content; For each training case of the plurality of training cases, analyzing the visual content associated with the training case to determine one or more characteristics of the visual content associated with the training case; attributing the first visual content to at least one, but not all, of the plurality of training examples using one or more characteristics of the first visual content and characteristics of the visual content associated with the plurality of training examples; determining whether at least one of the visual contents associated with the training examples of the first subgroup is associated with at least one first resource; and for each of said at least one first resource, updating a data record associated with said resource based on the attribution; A system configured to perform an action.

20. 1. A method for attributing generated visual content to training examples, the method comprising: receiving first visual content generated using a generative model, the generative model being a result of training a machine learning algorithm using a plurality of training cases, each training case of the plurality of training cases being associated with the visual content; determining one or more characteristics of the first visual content; For each training case of the plurality of training cases, analyzing the visual content associated with the training case to determine one or more characteristics of the visual content associated with the training case; attributing the first visual content to at least one, but not all, of the plurality of training examples using one or more characteristics of the first visual content and characteristics of the visual content associated with the plurality of training examples; determining whether at least one of the visual contents associated with the training examples of the first subgroup is associated with at least one first resource; and for each of said at least one first resource, updating a data record associated with said resource based on the attribution; The method according to claim 1, further comprising:

21. 1. A non-transitory computer-readable medium comprising instructions that cause at least one processor to perform operations to identify visual content utilized in training an inference model, the operations comprising: Receive distinctive visual content; accessing data based on at least one parameter of the inference model, the inference model being a result of training a machine learning algorithm using a plurality of training cases, each training case of the plurality of training cases may include the visual content; analyzing the data and the distinctive visual content to determine a likelihood that the distinctive visual content is included in at least one training example of the plurality of training examples; and generating a digital signal indicative of the likelihood that the distinctive visual content is included in at least one training example of the plurality of training examples. A non-transitory computer-readable medium comprising:

22. 22. The non-transitory computer-readable medium of claim 21 , wherein the likelihood is a binary likelihood, and determining the likelihood that the distinctive visual content is included in at least one training example of the plurality of training examples comprises determining whether the distinctive visual content is included in at least one training example of the plurality of training examples.

23. The operation further comprises: utilizing the particular visual content and the at least one parameter of the inference model to obtain an output of the inference model corresponding to the particular visual content; and the data is based on the output of the inference model corresponding to the distinctive visual content; 22. The non-transitory computer-readable medium of claim 21, comprising:

24. The operation further comprises: Utilizing the distinctive visual content to generate a plurality of variations of the distinctive visual content; for each variant of the plurality of variants, obtaining an output of the inference model corresponding to the variant using the variant and the at least one parameter of the inference model; and the data is based on the output of the inference model corresponding to the plurality of transformations; 22. The non-transitory computer-readable medium of claim 21, comprising:

25. The operation further comprises: Obtain multiple directions in mathematical space, obtaining distinctive mathematical objects in the mathematical space corresponding to the distinctive visual content; for each of the plurality of directions, using the unique mathematical object and the direction to determine a mathematical object in the mathematical space that corresponds to the unique mathematical object and the direction; and generating, for each of the plurality of directions, visual content corresponding to the unique mathematical object and the mathematical object in the mathematical space corresponding to the direction, thereby generating the plurality of variations of the unique visual content; 25. The non-transitory computer-readable medium of claim 24, comprising:

26. 26. The non-transitory computer-readable medium of claim 25, wherein the mathematical space is a non-orientable space.

27. The operation further comprises: selecting a plurality of objects, none of which is depicted within the distinctive visual content; and for each object of the plurality of objects, analyzing the distinctive visual content to generate a variant of the distinctive visual content that includes a depiction of the object, thereby generating the plurality of variants of the distinctive visual content; 25. The non-transitory computer-readable medium of claim 24, comprising:

28. The operation further comprises: analyzing the distinctive visual content to detect a plurality of objects depicted within the distinctive visual content; and for each object of the plurality of objects, analyzing the distinctive visual content to generate a variant of the distinctive visual content that does not include a depiction of the object, thereby generating the plurality of variants of the distinctive visual content; 25. The non-transitory computer-readable medium of claim 24, comprising:

29. The operation further comprises: Utilizing the characteristic visual content and the at least one parameter of the inference model to obtain a gradient corresponding to the characteristic visual content; and the data is based on the gradient corresponding to the particular visual content; 22. The non-transitory computer-readable medium of claim 21, comprising:

30. The operation further comprises: Utilizing the unique visual content and the at least one parameter of the inference model to calculate a loss function associated with the machine learning algorithm and a loss corresponding to the unique visual content; and the data is based on the loss function and the loss corresponding to the particular visual content; 22. The non-transitory computer-readable medium of claim 21, comprising:

31. The operation further comprises: analyzing the data and the distinctive visual content to determine a likelihood that an enhanced version of the distinctive visual content is included in at least one training example of the plurality of training examples; and generating a second digital signal indicative of the likelihood that the augmented version of the distinctive visual content is included in at least one training example of the plurality of training examples.

22. The non-transitory computer-readable medium of claim 21, comprising:

32. 22. The non-transitory computer-readable medium of claim 21 , wherein the inference model is a generative model, and for each training example of the plurality of training examples, the visual content included in the training example is a pair of visual content, an input visual content, and a desired output visual content.

33. The operation further comprises: calculating a convolution of at least a portion of the characteristic visual content, thereby obtaining a calculated convolution result value of at least a portion of the characteristic visual content; calculating a mathematical function of the resulting value of the calculated convolution of at least a portion of the distinctive visual content; selecting a threshold value based on said data; comparing the threshold value to the mathematical function of the result of the calculated convolution of at least a portion of the distinctive visual content; and determining a likelihood that the distinctive visual content is included in at least one training example of the plurality of training examples based on a result of the comparison of the threshold value and the mathematical function of the result of the calculated convolution of at least a portion of the distinctive visual content.

22. The non-transitory computer-readable medium of claim 21, comprising:

34. The operation further comprises: analyzing the data and the distinctive visual content to determine a similarity between the distinctive visual content and a selected training example of the plurality of training examples; and generating a second digital signal indicative of the degree of similarity between the distinctive visual content and the selected training example of the plurality of training examples.

22. The non-transitory computer-readable medium of claim 21, comprising:

35. 35. The non-transitory computer-readable medium of claim 34, wherein the similarity measures indicate an amount of expansion.

36. 35. The non-transitory computer readable medium of claim 34, wherein the second digital signal is indicative of the selected training example of the plurality of training examples.

37. 35. The non-transitory computer-readable medium of claim 34, wherein the selected training example of the plurality of training examples is a training example in the plurality of training examples that is most similar to the distinctive visual content.

38. The operation further comprises: accessing a data structure that associates visual content with resources to determine whether the particular visual content is associated with a particular resource; and indicating said unique resource in said generated digital signal.

22. The non-transitory computer-readable medium of claim 21, comprising:

39. 1. A system for identifying visual content used to train an inference model, the system comprising: at least one processor, the at least one processor comprising: Receive distinctive visual content; accessing data based on at least one parameter of the inference model, the inference model being a result of training a machine learning algorithm using a plurality of training cases, each training case of the plurality of training cases including visual content; analyzing the data and the distinctive visual content to determine a likelihood that the distinctive visual content is included in at least one training example of the plurality of training examples; and generating a digital signal indicative of the likelihood that the distinctive visual content is included in at least one training example of the plurality of training examples. A system configured to perform an action.

40. 1. A method for identifying visual content used to train an inference model, the method comprising: Receive distinctive visual content; accessing data based on at least one parameter of the inference model, the inference model being a result of training a machine learning algorithm using a plurality of training cases, each training case of the plurality of training cases including visual content; analyzing the data and the distinctive visual content to determine a likelihood that the distinctive visual content is included in at least one training example of the plurality of training examples; and generating a digital signal indicative of the likelihood that the distinctive visual content is included in at least one training example of the plurality of training examples. The method according to claim 1, further comprising:

41. 1. A non-transitory computer-readable medium comprising instructions that cause at least one processor to perform operations for transforming a non-photorealistic virtual environment into a photorealistic virtual environment, the operations comprising: receiving a first digital signal representing virtual content within an augmented reality environment; utilizing the first digital signal to identify non-photorealistic portions of the virtual content; analyzing the first digital signal utilizing a generative model to generate a photo-realistic version of the identified non-photo-realistic portion of the virtual content; and generating a second digital signal configured to cause a wearable augmented reality device to present the generated photo-realistic version of the virtual content in place of the identified non-photo-realistic portion of the virtual content within the augmented reality environment; A non-transitory computer-readable medium comprising:

42. 42. The non-transitory computer-readable medium of claim 41, wherein the identified non-photorealistic portion of the virtual content comprises a virtual room within the augmented reality environment.

43. 42. The non-transitory computer-readable medium of claim 41, wherein the identified non-photorealistic portions of the virtual content move within the augmented reality environment.

44. 42. The non-transitory computer-readable medium of claim 41 , wherein the identified non-photorealistic portions of the virtual content include an avatar associated with an individual, the photorealistic version of the identified non-photorealistic portions of the virtual content is a photorealistic depiction of the individual, and the analysis utilizing the generative model is based on a photograph of the individual.

45. 42. The non-transitory computer-readable medium of claim 41, wherein the virtual content includes a second non-photorealistic portion that is different from the identified non-photorealistic portion of the virtual content, and the second digital signal is configured to cause the wearable augmented reality device to present the generated photorealistic version of the virtual content in place of the identified non-photorealistic portion within the augmented reality environment while presenting the second non-photorealistic portion.

46. 42. The non-transitory computer-readable medium of claim 41, wherein the first digital signal is capable of indicating a field of view associated with the wearable augmented reality device, and the non-photorealistic portion of the virtual content is selected based on the field of view associated with the wearable augmented reality device.

47. 42. The non-transitory computer-readable medium of claim 41, wherein the first digital signal is a digital display signal configured to cause the wearable augmented reality device to present the virtual content within the augmented reality environment.

48. 42. The non-transitory computer-readable medium of claim 41, wherein the first digital signal encodes geometric data associated with the virtual content.

49. The operation further comprises: - calculating a convolution of at least a portion of said virtual content, thereby obtaining a calculated convolution result value of at least a portion of said virtual content; and utilizing the resulting values ​​of the computed convolution of at least a portion of the virtual content to identify the non-photorealistic portions of the virtual content.

42. The non-transitory computer-readable medium of claim 41, comprising:

50. 42. The non-transitory computer-readable medium of claim 41, wherein the operations further include utilizing a machine learning model to analyze the first digital signal and identify the non-photorealistic portions of the virtual content.

51. The operation further comprises: utilizing the first digital signal to cause the wearable augmented reality device to present the virtual content to the individual; and receiving input from the individual and identifying the non-photorealistic portions of the virtual content; 42. The non-transitory computer-readable medium of claim 41, comprising:

52. 42. The non-transitory computer-readable medium of claim 41, wherein the operations further include switching, based on input from an individual using the wearable augmented reality device, between causing the wearable augmented reality device to present the virtual content including the identified non-realistic portion of the virtual content within the augmented reality environment and causing the wearable augmented reality device to present the generated realistic version in place of the identified non-realistic portion of the virtual content.

53. The operation further comprises: analyzing the first digital signal to determine a mathematical object in a mathematical space that corresponds to the identified non-photorealistic portion of the virtual content; and generating the photo-realistic version of the identified non-photo-realistic portion of the virtual content utilizing the mathematical object in the mathematical space that corresponds to the identified non-photo-realistic portion of the virtual content; 42. The non-transitory computer-readable medium of claim 41, comprising:

54. 54. The non-transitory computer-readable medium of claim 53, wherein the mathematical space is a non-orientable space.

55. The operation further comprises: analyzing the first digital signal to determine a category of the non-photorealistic portion of the virtual content; and selecting a generative model from a plurality of generative models utilizing the category of the non-photorealistic portion of the virtual content; 42. The non-transitory computer-readable medium of claim 41, comprising:

56. 42. The non-transitory computer-readable medium of claim 41, wherein the non-photorealistic portions of the virtual content include non-photorealistic skin tones, and the photorealistic version of the identified non-photorealistic portions of the virtual content replaces the non-photorealistic skin tones with photorealistic skin tones.

57. 42. The non-transitory computer-readable medium of claim 41 , wherein the non-photorealistic portion of the virtual content includes a first element, a second element, and a non-photorealistic ratio between dimensions of the first element and dimensions of the second element, and generating the photorealistic version includes resizing at least one of the first element or the second element, and the photorealistic version of the identified non-photorealistic portion of the virtual content includes a photorealistic ratio between dimensions of the first element and dimensions of the second element.

58. 42. The non-transitory computer-readable medium of claim 41, wherein the non-photorealistic portion of the virtual content includes a non-photorealistic landscape, and the photorealistic version of the non-photorealistic portion of the virtual content replaces the non-photorealistic landscape with a photorealistic landscape.

59. 1. A system for causing at least one processor to perform operations for converting a non-photorealistic virtual environment into a photorealistic virtual environment, the system comprising: receiving a first digital signal representing virtual content within an augmented reality environment; utilizing the first digital signal to identify non-photorealistic portions of the virtual content; analyzing the first digital signal utilizing a generative model to generate a photo-realistic version of the identified non-photo-realistic portion of the virtual content; and generating a second digital signal configured to cause a wearable augmented reality device to present the generated photo-realistic version of the virtual content in place of the identified non-photo-realistic portion of the virtual content within the augmented reality environment; A system configured to perform an operation.

60. 16. A method for causing at least one processor to perform operations for converting a non-photorealistic virtual environment into a photorealistic virtual environment, the method comprising: receiving a first digital signal representing virtual content within an augmented reality environment; utilizing the first digital signal to identify non-photorealistic portions of the virtual content; analyzing the first digital signal utilizing a generative model to generate a photo-realistic version of the identified non-photo-realistic portion of the virtual content; and generating a second digital signal configured to cause a wearable augmented reality device to present the generated photo-realistic version of the virtual content in place of the identified non-photo-realistic portion of the virtual content within the augmented reality environment; The method according to claim 1, further comprising:

61. 1. A non-transitory computer-readable medium comprising instructions that cause at least one processor to perform operations for determining realism of artificially generated visual content, the operations comprising: accessing one or more synthetically-generated visual contents, the one or more synthetically-generated visual contents including a particular synthetically-generated visual content; accessing one or more captured visual contents, each visual content of the one or more captured visual contents being captured from a real physical environment using an image sensor; presenting, for each person of a plurality of people, a mix of a plurality of visual contents to the person, the mix of a plurality of visual contents including at least one visual content of the one or more artificially generated visual contents and at least one visual content of the one or more captured visual contents, the mix of a plurality of visual contents including the particular artificially generated visual content; For each person of the plurality of people, receiving a response to the presentation from the person, for each visual content of a group of at least one mix of visual content, the response indicating whether the person believed the visual content to be artificially generated visual content; and determining a realism of the particular artificially generated visual content based on the response; A non-transitory computer-readable medium comprising:

62. 62. The non-transitory computer-readable medium of claim 61, wherein the realism of the particular artificially-generated visual content is a relative degree to at least one other artificially-generated visual content.

63. 62. The non-transitory computer-readable medium of claim 61 , wherein the realism of the particular artificially-generated visual content is based on a comparison of the response to the particular artificially-generated visual content and at least one response of the one or more artificially-generated visual content to other artificially-generated visual content.

64. 62. The non-transitory computer-readable medium of claim 61, wherein the realism of the particular artificially-generated visual content is based on a comparison of the response to the particular artificially-generated visual content and at least one response to visual content of the one or more captured visual content.

65. 62. The non-transitory computer-readable medium of claim 61 , wherein the mix of visual content presented to one of the plurality of people is different from the mix of visual content presented to another of the plurality of people.

66. 62. The non-transitory computer-readable medium of claim 61 , wherein each artificially-generated visual content of the one or more artificially-generated visual contents is generated utilizing a conditional generative adversarial network having different input conditions.

67. 62. The non-transitory computer-readable medium of claim 61 , wherein the operations further include determining whether to remove the particular artificially-generated visual content based on the determined realism of the particular artificially-generated visual content.

68. 62. The non-transitory computer-readable medium of claim 61, wherein the operations further include determining whether to utilize the particular artificially-generated visual content for a particular application based on the determined realism of the particular artificially-generated visual content.

69. 62. The non-transitory computer-readable medium of claim 61 , wherein the operations further include analyzing the particular artificially-generated visual content and determining at least one reason for the determined realism of the particular artificially-generated visual content.

70. 70. The non-transitory computer-readable medium of claim 69, wherein the operations further include utilizing the at least one reason to generate new artificially generated visual content.

71. 70. The non-transitory computer-readable medium of claim 69, wherein the action further comprises presenting the determined at least one reason to an individual.

72. The operation further comprises: calculating a convolution of at least a portion of said particular artificially-generated visual content, thereby obtaining a calculated convolution result value of at least a portion of said particular artificially-generated visual content; and determining said at least one reason for the determined realism of said particular artificially-generated visual content using said resulting value of the calculated convolution of at least a portion of said particular artificially-generated visual content.

70. The non-transitory computer readable medium of claim 69, comprising:

73. the one or more artificially generated visual contents include a unique artificially generated visual content, the unique artificially generated visual content being different from the particular artificially generated visual content, and the operation further comprises: determining a realism of the distinctive artificially-generated visual content based on at least one response to the distinctive artificially-generated visual content; comparing the data associated with the unique artificially-generated visual content with the data associated with the particular artificially-generated visual content; and determining at least one reason for the determined realism of the particular artificially-generated visual content based on a result of the comparison; 62. The non-transitory computer-readable medium of claim 61, comprising:

74. 74. The non-transitory computer-readable medium of claim 73, wherein each visual content of the one or more artificially generated visual contents is the result of applying a function to different input values, the data associated with the particular artificially generated visual content being the input values ​​associated with the particular artificially generated visual content, and the data associated with the unique artificially generated visual content being the input values ​​associated with the unique artificially generated visual content.

75. The operation further comprises: analyzing the particular artificially-generated visual content to determine the data associated with the particular artificially-generated visual content; and analyzing the unique artificially-generated visual content to determine the data associated with the unique artificially-generated visual content; 74. The non-transitory computer-readable medium of claim 73, comprising:

76. The operation further comprises: calculating a convolution of at least a portion of said particular artificially-generated visual content, thereby determining said data related to said particular artificially-generated visual content; and computing a convolution of at least a portion of said distinctive artificially-generated visual content, thereby determining said data associated with said distinctive artificially-generated visual content; 74. The non-transitory computer-readable medium of claim 73, comprising:

77. the one or more artificially generated visual contents include a unique artificially generated visual content, the unique artificially generated visual content being different from the particular artificially generated visual content, and the operation further comprises: determining a realism of the distinctive artificially-generated visual content based on at least one response to the distinctive artificially-generated visual content; determining a particular mathematical object in said mathematical space that is associated with said particular artificially generated visual content; determining a unique mathematical object in the mathematical space that is associated with the unique artificially-generated visual content; determining a new mathematical object in the mathematical space using the determined realism of the particular artificially-generated visual content, the determined realism of the unique artificially-generated visual content, the particular mathematical object, and the unique mathematical object, wherein the new mathematical object is different from the particular mathematical object and the unique mathematical object; and generating new artificially generated visual content utilizing the new mathematical objects in the mathematical space; 62. The non-transitory computer-readable medium of claim 61, comprising:

78. The operation further comprises: For each visual content of the one or more artificially generated visual content, analyzing the visual content utilizing a machine learning model to determine a level of confidence that the visual content is realistic; and selecting the particular artificially-generated visual content from the one or more artificially-generated visual contents based on the level of certainty; 62. The non-transitory computer-readable medium of claim 61, comprising:

79. 1. A system for determining realism of artificially generated visual content, the system comprising at least one processor, the at least one processor comprising: accessing one or more synthetically-generated visual contents, the one or more synthetically-generated visual contents including a particular synthetically-generated visual content; accessing one or more captured visual contents, each visual content of the one or more captured visual contents being captured from a real physical environment using an image sensor; presenting, for each person of a plurality of people, a mix of a plurality of visual contents to the person, the mix of the plurality of visual contents including at least one visual content of the one or more artificially generated visual contents and at least one visual content of the one or more captured visual contents, the mix of the plurality of visual contents including the particular artificially generated visual content; For each person of the plurality of people, receiving a response to the presentation from the person, for each visual content of a group of at least one mix of visual content, the response indicating whether the person believed the visual content to be artificially generated visual content; and determining a realism of the particular artificially generated visual content based on the response; A system configured to perform an action.

80. 1. A method for determining realism of artificially generated visual content, the method comprising: accessing one or more synthetically-generated visual contents, the one or more synthetically-generated visual contents including a particular synthetically-generated visual content; accessing one or more captured visual contents, each visual content of the one or more captured visual contents being captured from a real physical environment using an image sensor; presenting, for each person of a plurality of people, a mix of a plurality of visual contents to the person, the mix of the plurality of visual contents including at least one visual content of the one or more artificially generated visual contents and at least one visual content of the one or more captured visual contents, the mix of the plurality of visual contents including the particular artificially generated visual content; For each person of the plurality of people, receiving a response to the presentation from the person, for each visual content of a group of at least one mix of visual content, the response indicating whether the person believed the visual content to be artificially generated visual content; and determining a realism of the particular artificially generated visual content based on the response; The method according to claim 1, further comprising:

Images