Field guided feature propagation for machine learning applications

Abstract

This disclosure provides systems, methods, and devices for machine learning techniques that support feature propagation. In one aspect, a method may include receiving data for processing by a first machine learning model, determining a propagation field identifies latent features within the data, and determining, with the first machine learning model, output data based on the propagation field and data. A feature propagation process of the first machine learning model may be guided by the latent features from the propagation field. Guiding the feature propagation process may include calculating initial propagation values that correspond to values of the propagation field. Additionally or alternatively, guiding the feature propagation process may include comparing segments of the propagation field against one or more thresholds and performing feature propagation for portions of the data that satisfy the one or more thresholds.

Description

FIELD GUIDED FEATURE PROPAGATION FOR MACHINE LEARNING APPLICATIONSCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 549,668, entitled, “FIELD GUIDED FEATURE PROPAGATION FOR MACHINE LEARNING APPLICATIONS,” filed on February 5, 2024, which is expressly incorporated by reference herein in its entirety.TECHNICAL FIELD

[0002] Aspects of the present disclosure relate generally to machine learning techniques, and more particularly, to methods and systems suitable for guiding feature propagation in machine learning applications using one or more fields.INTRODUCTION

[0003] Machine learning techniques encompass a diverse array of computational methodologies designed to enable systems to learn from and make predictions or decisions based on data. These techniques typically involve the construction of models, algorithms, or neural network architectures that can infer patterns, trends, or structures within large datasets without explicit programming for each task. Machine learning techniques include supervised learning, where models are trained using labeled datasets; unsupervised learning, which involves the identification of patterns in unlabeled data; semi-supervised learning, which combines both labeled and unlabeled data; and reinforcement learning, where models learn optimal behaviors through trial and error interactions with an environment.BRIEF SUMMARY OF SOME EXAMPLES

[0004] The following summarizes some aspects of the present disclosure to provide a basic understanding of the discussed technology. This summary is not an extensive overview of all contemplated features of the disclosure and is intended neither to identify key or critical elements of all aspects of the disclosure nor to delineate the scope of any or all aspects of the disclosure. Its sole purpose is to present some concepts of one or more aspects of the disclosure in summary form as a prelude to the more detailed description that is presented later.

[0005] One embodiment provides a method that includes receiving data for use by a first machine learning model, determining a propagation field for the data, wherein the field identifies latent features within the data, and determining, with the first machine learning model, output data based on the field and the data, wherein the first machine learning model performs at least one feature propagation process based on the latent features identified by the propagation field.

[0006] Another embodiment provides a system that includes at least one processor, at least one sensor, and a memory storing instructions which, when executed by the at least one processor, cause the at least one processor to receive, from the at least one sensor, data for use by a first machine learning model, determine a propagation field for the data, wherein the field identifies latent features within the data, and determine, with the first machine learning model, output data based on the field and the data, wherein the first machine learning model performs at least one feature propagation process based on the latent features identified by the propagation field.

[0007] A further embodiment provides a non-transitory, computer-readable medium storing instructions which, when executed by a processor, cause the processor to receive data for use by a first machine learning model, determine a propagation field for the data, wherein the field identifies latent features within the data, and determine, with the first machine learning model, output data based on the field and the data, wherein the first machine learning model performs at least one feature propagation process based on the latent features identified by the propagation field.

[0008] The foregoing has outlined rather broadly the features and technical advantages of examples according to the disclosure in order that the detailed description that follows may be better understood. Additional features and advantages will be described hereinafter. The conception and specific examples disclosed may be readily utilized as a basis for modifying or designing other structures for carrying out the same purposes of the present disclosure. Such equivalent constructions do not depart from the scope of the appended claims. Characteristics of the concepts disclosed herein, both their organization and method of operation, together with associated advantages will be better understood from the following description when considered in connection with the accompanying figures. Each of the figures is provided for the purposes of illustration and description, and not as a definition of the limits of the claims.

[0009] While aspects and implementations are described in this application by illustration to some examples, those skilled in the art will understand that additional implementationsand use cases may come about in many different arrangements and scenarios. Innovations described herein may be implemented across many differing platform types, devices, systems, shapes, sizes, packaging arrangements. For example, implementations or uses may come about via integrated chip implementations or other non-module-component based devices (e.g., end-user devices, vehicles, communication devices, computing devices, industrial equipment, retail devices or purchasing devices, medical devices, AI- enabled devices, etc.). While some examples may or may not be specifically directed to use cases or applications, a wide assortment of applicability of described innovations may occur.

[0010] Implementations may range from chip-level or modular components to non-modular, non-chip-level implementations and further to aggregated, distributed, or original equipment manufacturer (OEM) devices or systems incorporating one or more described aspects. In some practical settings, devices incorporating described aspects and features may also necessarily include additional components and features for implementation and practice of claimed and described aspects. It is intended that innovations described herein may be practiced in a wide variety of implementations, including both large devices or small devices, chip-level components, multi-component systems (e.g., radio frequency (RF)-chain, communication interface, processor), distributed arrangements, end-user devices, etc. of varying sizes, shapes, and constitution.

[0011] In the following description, numerous specific details are set forth, such as examples of specific components, circuits, and processes to provide a thorough understanding of the present disclosure. The term “coupled” as used herein means connected directly to or connected through one or more intervening components or circuits. Also, in the following description and for purposes of explanation, specific nomenclature is set forth to provide a thorough understanding of the present disclosure. However, it will be apparent to one skilled in the art that these specific details may not be required to practice the teachings disclosed herein. In other instances, well known circuits and devices are shown in block diagram form to avoid obscuring teachings of the present disclosure.

[0012] Some portions of the detailed descriptions which follow are presented in terms of procedures, logic blocks, processing, and other symbolic representations of operations on data bits within a computer memory. In the present disclosure, a procedure, logic block, process, or the like, is conceived to be a self-consistent sequence of steps or instructions leading to a desired result. The steps are those requiring physical manipulations of physical quantities. Usually, although not necessarily, these quantities take the form ofelectrical or magnetic signals capable of being stored, transferred, combined, compared, and otherwise manipulated in a computer system.

[0013] In the figures, a single block may be described as performing a function or functions. The function or functions performed by that block may be performed in a single component or across multiple components, and / or may be performed using hardware, software, or a combination of hardware and software. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps are described below generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure. Also, the example devices may include components other than those shown, including well-known components such as a processor, memory, and the like.

[0014] Unless specifically stated otherwise as apparent from the following discussions, it is appreciated that throughout the present application, discussions utilizing the terms such as “accessing,” “receiving,” “sending,” “using,” “selecting,” “determining,” “normalizing,” “multiplying,” “averaging,” “monitoring,” “comparing,” “applying,” “updating,” “measuring,” “deriving,” “settling,” “generating” or the like, refer to the actions and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system’s registers and memories into other data similarly represented as physical quantities within the computer system’s registers, memories, or other such information storage, transmission, or display devices.

[0015] The terms “device” and “apparatus” are not limited to one or a specific number of physical objects (such as one smartphone, one camera controller, one processing system, and so on). As used herein, a device may be any electronic device with one or more parts that may implement at least some portions of the disclosure. While the below description and examples use the term “device” to describe various aspects of the disclosure, the term “device” is not limited to a specific configuration, type, or number of objects. As used herein, an apparatus may include a device or a portion of the device for performing the described operations.

[0016] As used herein, including in the claims, the term “or,” when used in a list of two or more items, means that any one of the listed items may be employed by itself, or any combination of two or more of the listed items may be employed. For example, if a composition is described as containing components A, B, or C, the composition may contain A alone; B alone; C alone; A and B in combination; A and C in combination; B and C in combination; or A, B, and C in combination.

[0017] Also, as used herein, including in the claims, “or” as used in a list of items prefaced by “at least one of’ indicates a disjunctive list such that, for example, a list of “at least one of A, B, or C” means A or B or C or AB or AC or BC or ABC (that is A and B and C) or any of these in any combination thereof.

[0018] Also, as used herein, the term “substantially” is defined as largely but not necessarily wholly what is specified (and includes what is specified; for example, substantially 90 degrees includes 90 degrees and substantially parallel includes parallel), as understood by a person of ordinary skill in the art. In any disclosed implementations, the term “substantially” may be substituted with “within [a percentage] of’ what is specified, where the percentage includes .1, 1, 5, or 10 percent.

[0019] Also, as used herein, relative terms, unless otherwise specified, may be understood to be relative to a reference by a certain amount. For example, terms such as “higher” or “lower” or “more” or “less” may be understood as higher, lower, more, or less than a reference value by a threshold amount.BRIEF DESCRIPTION OF THE DRAWINGS

[0020] A further understanding of the nature and advantages of the present disclosure may be realized by reference to the following drawings. In the appended figures, similar components or features may have the same reference label. Further, various components of the same type may be distinguished by following the reference label by a dash and a second label that distinguishes among the similar components. If just the first reference label is used in the specification, the description is applicable to any one of the similar components having the same first reference label irrespective of the second reference label.

[0021] FIG. 1 shows a block diagram of an example image processing configuration for a vehicle according to one or more aspects of the disclosure.

[0022] FIG. 2 is a block diagram illustrating details of an example wireless communication system according to one or more aspects.

[0023] FIG. 3 is a block diagram illustrating a system for performing field guided feature propagation according to one or more aspects of the disclosure.

[0024] FIG. 4 is a flow chart illustrating an example method for performing field guided feature propagation according to one or more aspects of the disclosure.

[0025] FIGs. 5A-5B show example fields according to aspects of the disclosure.

[0026] Like reference numbers and designations in the various drawings indicate like elements.DETAILED DESCRIPTION

[0027] The detailed description set forth below, in connection with the appended drawings, is intended as a description of various configurations and is not intended to limit the scope of the disclosure. Rather, the detailed description includes specific details for the purpose of providing a thorough understanding of the inventive subject matter. It will be apparent to those skilled in the art that these specific details are not required in every case and that, in some instances, well-known structures and components are shown in block diagram form for clarity of presentation.

[0028] The present disclosure provides systems, apparatus, methods, and computer-readable media that support improved feature propagation for machine learning models. Feature propagation in machine learning models, such as neural networks, is a technique used to refine estimates of target features through successive iterations. The general idea is to derive more accurate mappings of features by measuring their similarities across different instances within a dataset, such as frames in a video or corresponding points in stereo image pairs. Feature propagation is critical for tasks that require a high level of precision, such as image reconstruction, 3D modeling, and object tracking.

[0029] In neural networks, feature propagation often involves the use of model layers (such as convolutional layers, transformer modules, other types of neural operations, and the like) to process and pass feature information throughout the network architecture. As data moves through the network, each layer extracts and propagates higher-level feature representations built upon the outputs of the previous layers, ultimately leading to a robust understanding of the data.

[0030] One exemplary process for feature propagation (such as in computer vision tasks) is the PatchMatch process. PatchMatch is designed for estimating correspondences between patches in two or more received data sets (such as two or more images, two or more regions of features, two or more sets of sensor measurements, and the like) and may typically operate in three main phases: (1) an initialization phase, (2) a propagation phase,and (3) a search phase. During the initialization phase, the process begins by assigning a random estimate of the target feature for each unit of analysis, which could be individual measurements or multiple measurements. For example, in the context of calculating disparity for stereo image pairs, the process may randomly assign a disparity value to each unit without considering its actual position in the corresponding data. As another example, the initialization step may randomly select corresponding pixels within stereo image pairs. During the propagation phase, the estimates are propagated to neighboring units according to computed similarity scores. This propagation may be done in a rasterscan order, where each unit's current estimate is compared with the estimates of its neighbors. If a neighbor has a better (higher similarity) estimate, its estimate may be adopted. This process leverages the assumption that adjacent measurements are likely to have similar disparity values, encouraging spatial coherence in the estimated feature map. During the search phase, a local search is performed to escape local minima and approach a global solution. The search explores a wider range around the current estimate to find an improved match that may not have been considered during the narrower propagation phase. This is particularly useful for overcoming the local minima issue and ensuring that the disparity estimates are as close to the global optimum as possible.

[0031] However, existing feature propagation techniques may result in significant inaccuracies in certain scenarios. For example, when using machine learning models to find similarities within images (such as for stereo imaging applications), accuracy may be significantly degraded when the images include extensive areas that are textureless, or that contain repeating patterns. In particular, in stereo matching tasks, the goal is to match corresponding features between a pair of input images to determine depth information. However, when surfaces lack texture or distinct features — for instance, a uniformly painted wall or a smooth table — existing techniques struggle to establish correct correspondences because there are insufficient unique visual clues to differentiate one area from another. Incorrectly identifying corresponding portions of images may cause features to propagate incorrectly (such as by propagating features for a background of the image to the foreground of the image), which may result in incorrect output (such as incorrect depth information). Such situations lead to errors that not only diminish the accuracy of depth estimation but can also produce wider safety implications, especially in dynamic, real-world applications where reliable depth cues are crucial for navigation and object interaction.

[0032] As another example, random initialization can hinder the overall performance and effectiveness of feature propgation techniques in certain scenarios. Methods like the PatchMatch process often rely on random initialization as a starting point for estimating feature correspondences between images. These random initializations can result in or require many iterations (e.g., of searching, propagation, or a combination thereof) to converge on optimal results for a particular feature. Thus, such initializations can result in longer search durations and increased iteration counts to reach convergence, as the process must work through unnecessary computational steps to correct for the arbitrary beginning. The reliance on randomness not only inflates the complexity and computational costs but also presents an opportunity for errors to be introduced earlier in the process, possibly affecting the accuracy and resolution of the final feature mapping.

[0033] Additionally, error propagation may be an issue within existing techniques for feature propagation. In particular, when an error is introduced in the early stages of feature matching (such as due to an inability to deal with weak or indistinct features, such as those found on textureless surfaces), the error can be amplified through subsequent iterations. One example is an initially incorrect estimate of a feature's position or correspondence, which, instead of being corrected in later iterations, is compounded as the feature propagation process builds upon this flawed foundation, leading to increasingly inaccurate results. This cascading effect of errors, known as error propagation, undermines the integrity of the final outcome, rendering the process unreliable. Error propagation may be particularly problematic in scenarios where precision is paramount, as any initial inaccuracy can escalate and result in substantial deviations from the true values or states being estimated.

[0034] One solution to these problems is to initially determine a propagation field based on the data and to use that field to guide the feature propagation. Different techniques for guiding the feature propagation are discussed in greater detail below. In one example, to reduce the problems associated random initialization, an epistemic field may be determined based on previous values for features, and may be used as initialization values for propagation. Such fields may serve as an initial guess informed by previous observations, which can be more accurate than the random selections of traditional PatchMatch processes. For instance, if sequential frames of a moving car are being processed, knowledge of the car's consistent trajectory can provide a less random and more probable initialization. Consequently, this reduces the number of iterations necessary to reach convergence, which reduces processing time and latency, allowing forfaster analysis and improved model performance. This may also mitigate error propagation by decreasing the opportunities for errors to accumulate and propagate through each iteration. In another example, fields may be used to restrict feature propagation within particular regions of data (such as within the foreground / background of an image). Such fields may be particularly beneficial in reducing feature propagation errors caused by regions where features are indistinct, as it maintains the propagation of features within corresponding portions of the data.

[0035] Different types of fields are discussed in greater detail below. Semantic fields, one of these types, may include guiding masks or other indications of semantic information for received data, such as by segmenting the image according to one or more categories (e.g., foreground / background, detected objects, and the like). Epistemic fields may be determined based on knowledge or models of the physical or expected processes within the data, offering predictions for feature states. They may provide an informed starting point for propagation that's grounded in the dynamics captured in historical data, assisting in overcoming the randomness of traditional initializations. These predictions can range from simple linear extrapolations to complex formulations derived from physical laws or historical behavior patterns. In further implementations, propagation fields may incorporate input from external sources, such as user commands or real-world sensors, to guide propagation according to user-defined criteria or environmental feedback. This integration of external cues into the propagation process can enable models to adaptively refine feature estimates based on real-time prompts or additional contextual information that falls outside the purview of the immediate sensory data, offering dynamic, customizable, and interactive feature estimation and model performance.

[0036] Particular implementations of the subject matter described in this disclosure may be implemented to realize one or more of the following potential advantages or benefits. The described techniques provide significant benefits by enhancing the precision and efficiency of feature propagation in computer vision. Through the use of fields to guide the propagation process, the present techniques notably reduce the randomness in initialization, thereby limiting error propagation and streamlining convergence to accurate estimates. Furthermore, semantic fields allow for context-specific propagation, effectively compartmentalizing areas of interest and avoiding the pitfalls associated with indistinct features, such as those found on textureless surfaces. The employment of epistemic fields introduces an element of predictive modeling, offering a more robust and informed basis for initializing the feature estimation process that aligns with observedhistorical trends and physical behaviors. Additionally, the ability to incorporate fields based on extrinsic data opens up avenues for interactive and dynamic adjustments to the propagation process, accommodating real-time inputs and / or other data sources to improve or refine the feature propagation processes to focus on particular portions of the data. Collectively, these techniques provide a more holistically informed, adaptable, and reliable approach to feature propagation, resulting in improved accuracy, reduced computation overhead, and increased safety in applications reliant upon robust computer vision solutions. Furthermore, by improving the initialization values, these techniques may reduce the computing resources required to perform feature propagation.

[0037] FIG. 1 shows a block diagram of an example processing system 200 according to one or more aspects of the disclosure. The processing system 200 may include, or otherwise be coupled to, an image signal processor 212 for processing image frames from one or more image sensors, such as a first image sensor 201, a second image sensor 202, and a depth sensor 240. In some implementations, the processing system 200 also includes or is coupled to a processor (e.g., CPU) 204 and a memory 206 storing instructions 208. The processing system 200 may also include or be coupled to a display 214 and input / output (VO) components 216. I / O components 216 may be used for interacting with a user, such as a touch screen interface and / or physical buttons. I / O components 216 may also include network interfaces for communicating with other devices, such as other computing devices, mobile devices, vehicles, and / or a remote monitoring system. The network interfaces may include one or more of a wide area network (WAN) adaptor 252, a local area network (LAN) adaptor 253, and / or a personal area network (PAN) adaptor 254. An example WAN adaptor 252 is a 4G LTE or a 5G NR wireless network adaptor. An example LAN adaptor 253 is an IEEE 802.11 WiFi wireless network adapter. An example PAN adaptor 254 is a Bluetooth wireless network adaptor. Each of the adaptors 252, 253, and / or 254 may be coupled to an antenna, including multiple antennas configured for primary and diversity reception and / or configured for receiving specific frequency bands. The processing system 200 may further include or be coupled to a power supply 218, such as a mains power supply, a battery, and the like. The processing system 200 may also include or be coupled to additional features or components that are not shown in Figure 2. In one example, a wireless interface, which may include one or more transceivers and associated baseband processors, may be coupled to or included in WAN adaptor 252 for a wireless communication device. In a further example, an analogfront end (AFE) to convert analog image frame data to digital image frame data may be coupled between the image sensors 201 and 202 and the image signal processor 212.

[0038] The processing system 200 may include a sensor hub 250 for interfacing with and / or receiving data from sensors (such as non-camera sensors). One example non-camera sensor is a gyroscope, a device configured for measuring rotation, orientation, and / or angular velocity to generate motion data. Another example non-camera sensor is an accelerometer, a device configured for measuring acceleration, which may also be used to determine velocity and distance traveled by appropriately integrating the measured acceleration, and one or more of the acceleration, velocity, and or distance may be included in generated motion data. In further examples, a non-camera sensor may be a global positioning system (GPS) receiver, a light detection and ranging (LiDAR) system, a radio detection and ranging (RADAR) system, or other ranging systems. For example, the sensor hub 250 may interface to a vehicle bus for sending configuration commands and / or receiving information from vehicle sensors 272, such as distance (e.g., ranging) sensors or vehi cl e-to- vehicle (V2V) sensors (e.g., sensors for receiving information from nearby vehicles).

[0039] The image signal processor (ISP) 212 may receive image data, such as used to form image frames. In one embodiment, a local bus connection couples the image signal processor 212 to image sensors 201 and 202 of a first camera 203 and second camera 205, respectively. In another embodiment, a wire interface may couple the image signal processor 212 to an external image sensor. In a further embodiment, a wireless interface may couple the image signal processor 212 to the image sensor 201, 202.

[0040] The first camera 203 may include the first image sensor 201 and a corresponding first lens 231. The second camera 205 may include the second image sensor 202 and a corresponding second lens 232. Each of the lenses 231 and 232 may be controlled by an associated autofocus (AF) algorithm 233 executing in the ISP 212, which adjust the lenses 231 and 232 to focus on a particular focal plane at a certain scene depth from the image sensors 201 and 202. The AF algorithm 233 may be assisted by depth sensor 240. In some embodiments, the lenses 231 and 232 may have a fixed focus.

[0041] The first image sensor 201 and the second image sensor 202 are configured to capture one or more image frames. Lenses 231 and 232 focus light at the image sensors 201 and 202, respectively, through one or more apertures for receiving light, one or more shutters for blocking light when outside an exposure window, one or more color filter arrays (CFAs) for filtering light outside of specific frequency ranges, one or more analog frontends for converting analog measurements to digital information, and / or other suitable components for imaging.

[0042] Each of the cameras 203, 205 may include one, two, or more image sensors 201, 202. For example, the camera 203 may include a first image sensor 201 and a second image sensor (not depicted). When multiple image sensors are present, the first image sensor 201 may have a larger field of view (FOV) than the second image sensor or the first image sensor 201 may have different sensitivity or different dynamic range than the second image sensor. In one example, the first image sensor 201 may be a wide-angle image sensor, and the second image sensor may be a telephoto image sensor. In another example, the first image sensor 201 is configured to obtain an image through a first lens with a first optical axis and the second sensor is configured to obtain an image through a second lens with a second optical axis different from the first optical axis. Additionally or alternatively, the first lens may have a first magnification, and the second lens may have a second magnification different from the first magnification. This configuration may occur in a camera module with a lens cluster, in which the multiple image sensors and associated lenses are located in offset locations within the camera module. Additional image sensors may be included with larger, smaller, or same fields of view. Although the example discussed above focused on the first camera 203, the second camera 205 may be configured using one or more of the configurations discussed above (such as with a first image sensor 202 and a second image sensor (not depicted)).

[0043] Each image sensor may include means for capturing data representative of a scene, such as image sensors (including charge-coupled devices (CCDs), Bayer-filter sensors, infrared (IR) detectors, ultraviolet (UV) detectors, complimentary metal-oxide- semiconductor (CMOS) sensors), and / or time of flight detectors. The apparatus may further include one or more means for accumulating and / or focusing light rays into the one or more image sensors (including simple lenses, compound lenses, spherical lenses, and non-spherical lenses). These components may be controlled to capture the first, second, and / or more image frames. The image frames may be processed to form a single output image frame, such as through a fusion operation, and that output image frame further processed according to the aspects described herein.

[0044] As used herein, image sensor may refer to the image sensor itself and any certain other components coupled to the image sensor used to generate an image frame for processing by the image signal processor or other logic circuitry or storage in memory, whether a short-term buffer or longer-term non-volatile memory. For example, an image sensormay include other components of a camera, including a shutter, buffer, or other readout circuitry for accessing individual pixels of an image sensor. The image sensor may further refer to an analog front end or other circuitry for converting analog signals to digital representations for the image frame that are provided to digital circuitry coupled to the image sensor.

[0045] In some embodiments, the image signal processor 212 may execute instructions from a memory, such as instructions 208 from the memory 206, instructions stored in a separate memory coupled to or included in the image signal processor 212, or instructions provided by the processor 204. In addition, or in the alternative, the image signal processor 212 may include specific hardware (such as one or more integrated circuits (ICs)) configured to perform one or more operations described in the present disclosure. For example, the image signal processor 212 may include one or more image front ends (IFEs) 235, one or more image post-processing engines (IPEs) 236, and or one or more auto exposure compensation (AEC) 234 engines. The AF 233, AEC 234, IFE 235, IPE 236 may each include application-specific circuitry, be embodied as software code executed by the ISP 212, and / or a combination of hardware within and software code executing on the ISP 212.

[0046] In some implementations, the memory 206 may include a non-transient or non-transitory computer readable medium storing computer-executable instructions 208 to perform all or a portion of one or more operations described in this disclosure. In some implementations, the instructions 208 include a camera application (or other suitable application) to be executed during operation of the processing system 200 for generating images or videos. The instructions 208 may also include other applications or programs executed for the processing system 200, such as an operating system, mapping applications, or entertainment applications. Execution of the camera application, such as by the processor 204, may cause the processing system 200 to generate images using the image sensors 201 and 202 and the image signal processor 212. The memory 206 may also be accessed by the image signal processor 212 to store processed frames or may be accessed by the processor 204 to obtain the processed frames. In some embodiments, the processing system 200 includes a system on chip (SoC) that incorporates the image signal processor 212, the processor 204, the sensor hub 250, the memory 206, and input / output components 216 into a single package.

[0047] In some embodiments, at least one of the image signal processor 212 or the processor 204 executes instructions to perform various operations described herein, including objectdetection, image processing, natural language processing, text generation, risk map generation, driver monitoring, driver alert operations, and the like. For example, execution of the instructions can instruct the image signal processor 212 to begin or end capturing an image frame or a sequence of image frames. In some embodiments, the processor 204 may include one or more general-purpose processor cores 204A capable of executing scripts or instructions of one or more software programs, such as instructions 208 stored within the memory 206. For example, the processor 204 may include one or more application processors configured to execute the camera application (or other suitable application for generating images or video) stored in the memory 206. In executing the camera application, the processor 204 may be configured to instruct the image signal processor 212 to perform one or more operations with reference to the image sensors 201, 202, as discussed above.

[0048] In some embodiments, the processor 204 may include ICs or other hardware (e.g., an artificial intelligence (Al) engine 224) in addition to the ability to execute software to cause the processing system 200 to perform a number of functions or operations, such as the operations described herein. In some other embodiments, the processing system 200 does not include the processor 204, such as when all of the described functionality is configured in the image signal processor 212. In particular embodiments, the processor 204 and / or another processor of the processing system 200 may include a machine learning processor. Machine learning processors may include one or more processing units tailored for operating / manipulating machine learning data / features structures (e.g., tensors), executing machine learning algorithms, or a combination thereof. A first example machine learning processor includes Neural Processors (NPs), hardware components specifically designed to perform calculations necessary for artificial neural networks, leveraging parallel processing capabilities to handle complex computational tasks efficiently. A second example machine learning processor includes Hardware- Based Machine Learning Accelerators (MLAs) that enhance the speed of machine learning applications by optimizing the underlying hardware for specific machine learning algorithms (such as for particular types of computing operations). A third example machine learning processor may include an machine learning (ML) core within a CPU, which may be embedded in a traditional CPU and may be specifically optimized to accelerate machine learning workloads or computations. A fourth example machine learning processor may include Neural Signal Processors (NSPs) and / or NeuralProcessing Units (NPUs) are other types of processors that are designed for optimized performance with neural network-based workloads.

[0049] In some embodiments, the display 214 may include one or more suitable displays or screens allowing for user interaction and / or to present items to the user, such as a preview of the image frames being captured by the image sensors 201 and 202. In some embodiments, the display 214 is a touch-sensitive display. The I / O components 216 may be or include any suitable mechanism, interface, or device to receive input (such as commands) from the user and to provide output to the user through the display 214. For example, the I / O components 216 may include (but are not limited to) a graphical user interface (GUI), a keyboard, a mouse, a microphone, speakers, a squeezable bezel, one or more buttons (such as a power button), a slider, a switch, and so on.

[0050] While shown to be coupled to each other via the processor 204, components (such as the processor 204, the memory 206, the image signal processor 212, the display 214, and the I / O components 216) may be coupled to each another in other various arrangements, such as via one or more local buses, which are not shown for simplicity. While the image signal processor 212 is illustrated as separate from the processor 204, the image signal processor 212 may be a core of a processor 204 that is an application processor unit (APU), included in a system on chip (SoC), or otherwise included with the processor 204. While the processing system 200 is referred to in the examples herein for including aspects of the present disclosure, some device components may not be shown in Figure 2 to prevent obscuring aspects of the present disclosure. Additionally, other components, numbers of components, or combinations of components may be included in a suitable vehicle for performing aspects of the present disclosure. As such, the present disclosure is not limited to a specific device or configuration of components, including the processing system 200.

[0051] The processing system 200 may communicate as a user equipment (UE) within a wireless network 300, such as through WAN adaptor 252, as shown in FIG. 2. FIG. 2 is a block diagram illustrating details of an example wireless communication system according to one or more aspects. Wireless network 300 may, for example, include a 5G wireless network. As appreciated by those skilled in the art, components appearing in FIG. 2 are likely to have related counterparts in other network arrangements including, for example, cellular-style network arrangements and non-cellular-style-network arrangements (e.g., device-to-device or peer-to-peer or ad-hoc network arrangements, etc.).

[0052] Wireless network 300 includes base stations 305 and other network entities. A base station may be a station that communicates with the UEs and may also be referred to as an evolved node B (eNB), a next generation eNB (gNB), an access point, and the like. Each base station 305 may provide communication coverage for a particular geographic area. In 3GPP, the term “cell” may refer to this particular geographic coverage area of a base station or a base station subsystem serving the coverage area, depending on the context in which the term is used. In implementations of wireless network 300 herein, base stations 305 may be associated with a same operator or different operators (e.g., wireless network 300 may include a plurality of operator wireless networks). Additionally, in implementations of wireless network 300 herein, base station 305 may provide wireless communications using one or more of the same frequencies (e.g., one or more frequency bands in licensed spectrum, unlicensed spectrum, or a combination thereof) as a neighboring cell. In some examples, an individual base station 305 or UE 315 may be operated by more than one network operating entity. In some other examples, each base station 305 and UE 315 may be operated by a single network operating entity.

[0053] A base station may provide communication coverage for a macro cell or a small cell, such as a pico cell or a femto cell, or other types of cell. A macro cell generally covers a relatively large geographic area (e.g., several kilometers in radius) and may allow unrestricted access by UEs with service subscriptions with the network provider. A small cell, such as a pico cell, would generally cover a relatively smaller geographic area and may allow unrestricted access by UEs with service subscriptions with the network provider. A small cell, such as a femto cell, would also generally cover a relatively small geographic area (e.g., a home) and, in addition to unrestricted access, may also provide restricted access by UEs having an association with the femto cell (e.g., UEs in a closed subscriber group (CSG), UEs for users in the home, and the like). A base station for a macro cell may be referred to as a macro base station. A base station for a small cell may be referred to as a small cell base station, a pico base station, a femto base station or a home base station. In the example shown in FIG. 3, base stations 305d and 305e are regular macro base stations, while base stations 305a-305c are macro base stations enabled with one of three-dimension (3D), full dimension (FD), or massive MIMO. Base stations 305a-305c take advantage of their higher dimension MIMO capabilities to exploit 3D beamforming in both elevation and azimuth beamforming to increase coverage and capacity. Base station 305f is a small cell base station which may be a home node orportable access point. A base station may support one or multiple (e.g., two, three, four, and the like) cells.

[0054] Wireless network 300 may support synchronous or asynchronous operation. For synchronous operation, the base stations may have similar frame timing, and transmissions from different base stations may be approximately aligned in time. For asynchronous operation, the base stations may have different frame timing, and transmissions from different base stations may not be aligned in time. In some scenarios, networks may be enabled or configured to handle dynamic switching between synchronous or asynchronous operations.

[0055] UEs 315 are dispersed throughout the wireless network 300, and each UE may be stationary or mobile. It should be appreciated that, although a mobile apparatus is commonly referred to as a UE in standards and specifications promulgated by the 3 GPP, such apparatus may additionally or otherwise be referred to by those skilled in the art as a mobile station (MS), a subscriber station, a mobile unit, a subscriber unit, a wireless unit, a remote unit, a mobile device, a wireless device, a wireless communications device, a remote device, a mobile subscriber station, an access terminal (AT), a mobile terminal, a wireless terminal, a remote terminal, a handset, a terminal, a user agent, a mobile client, a client, a gaming device, an augmented reality device, vehicular component, vehicular device, or vehicular module, or some other suitable terminology.

[0056] Some non-limiting examples of a mobile apparatus, such as may include implementations of one or more of UEs 315, include a mobile, a cellular (cell) phone, a smart phone, a session initiation protocol (SIP) phone, a wireless local loop (WLL) station, a laptop, a personal computer (PC), a notebook, a netbook, a smart book, a tablet, a personal digital assistant (PDA), and a vehicle. Although UEs 315i-k are specifically shown as vehicles, a vehicle may employ the communication configuration described with reference to any of the UEs 315a-315k.

[0057] In one aspect, a UE may be a device that includes a Universal Integrated Circuit Card (UICC). In another aspect, a UE may be a device that does not include a UICC. In some aspects, UEs that do not include UICCs may also be referred to as loE devices. UEs 315a-315d of the implementation illustrated in FIG. 3 are examples of mobile smart phone-type devices accessing wireless network 300. A UE may also be a machine specifically configured for connected communication, including machine type communication (MTC), enhanced MTC (eMTC), narrowband loT (NB-IoT) and the like.UEs 315e-315k illustrated in FIG. 3 are examples of various machines configured for communication that access wireless network 300.

[0058] A mobile apparatus, such as UEs 315, may be able to communicate with any type of the base stations, whether macro base stations, pico base stations, femto base stations, relays, and the like. In FIG. 3, a communication link (represented as a lightning bolt) indicates wireless transmissions between a UE and a serving base station, which is a base station designated to serve the UE on the downlink or uplink, or desired transmission between base stations, and backhaul transmissions between base stations. UEs may operate as base stations or other network nodes in some scenarios. Backhaul communication between base stations of wireless network 300 may occur using wired or wireless communication links.

[0059] In operation at wireless network 300, base stations 305a-305c serve UEs 315a and 315b using 3D beamforming and coordinated spatial techniques, such as coordinated multipoint (CoMP) or multi-connectivity. Macro base station 305d performs backhaul communications with base stations 305a-305c, as well as small cell, base station 305f. Macro base station 305d also transmits multicast services which are subscribed to and received by UEs 315c and 315d. Such multicast services may include mobile television or stream video, or may include other services for providing community information, such as weather emergencies or alerts, such as Amber alerts or gray alerts.

[0060] Wireless network 300 of implementations supports mission critical communications with ultra-reliable and redundant links for mission critical devices, such UE 315e, which is a drone. Redundant communication links with UE 315e include from macro base stations 305d and 305e, as well as small cell base station 305f. Other machine type devices, such as UE 315f (thermometer), UE 315g (smart meter), and UE 315h (wearable device) may communicate through wireless network 300 either directly with base stations, such as small cell base station 305f, and macro base station 305e, or in multi-hop configurations by communicating with another user device which relays its information to the network, such as UE 315f communicating temperature measurement information to the smart meter, UE 315g, which is then reported to the network through small cell base station 305f. Wireless network 300 may also provide additional network efficiency through dynamic, low-latency TDD communications or low-latency FDD communications, such as in a vehicle-to-vehicle (V2V) mesh network between UEs 315i-315k communicating with macro base station 305e.

[0061] Aspects of the systems described with reference to, and shown in, FIGs. 1 and 2 may include determining propagation fields to guide feature propagation for machine learning models. In particular, attention weights may be determined to reduce or remove multiplication steps.

[0062] FIG. 3 is a block diagram illustrating a system 100 for field guided feature propagation. The system 100 may be an exemplary implementation of the processing system 200. The system 100 includes data 104, which is received by a computing device 102, which may be an exemplary implementation of the processing system 200. The computing device 102 includes a selection mask 106, a propagation field 108, a first machine learning model 110, a second machine learning model 112, a threshold 118, an output data 120. The first machine learning model 110 includes a feature propagation process 114, which includes initialization values 116.

[0063] The computing device 102 may be configured to receive data 104 for use by a first machine learning model 110. The computing device 102 is configured to receive various types of data 104 in certain implementations. Image data may consist of still photographs or frames, captured through an image sensor. Video data may include a series of temporally sequenced image frames that, when processed. Sensor data includes a broad range of inputs from environmental sensors (such as temperature, humidity, or pressure sensors), biometric sensors (such as capturing heart rate or fingerprint information), internet of things (loT) devices, and the like. Location data may include locations derived from Global Positioning System (GPS) or other location-tracking technologies. Position data may include point clouds or other position measurements for objects, such as using LIDAR or similar position sensing technology. These types of data 104 may be directly sourced from the corresponding sensors in real time or may be retrieved from pre-existing stores within databases, depending on the implementation.

[0064] The computing device 102 may be configured to determine a propagation field 108 for the data 104, the field identifies latent features within the data 104. Latent features within data 104 may refer to the inherent, non-explicit characteristics that are potentially extractable through analytical processes and may be relevant to downstream processing of the data 104, such as feature propagation. Latent features may include semantic features, epistemic features, and the like. Semantic features may represent latent meanings and contextual significance within the data, such as the possible subject matter of an image or the intended meaning behind a series of sensor readings. Epistemic features may be derived based on historical feature values, potentially embodying knowledgeaccumulated over time, which could indicate trends, anomalies, or provide predictive insights to inform future data interpretation. Additional example latent features may include abstract representations such as embeddings learned from text data, which may capture syntactic and semantic relationships between words, or latent variables in sensor data that might infer the operating state of a machine. Latent features may also include temporal features embedded within sequential data, including possible time series momentum or seasonality components, and spatial features that might capture the relative positions and arrangements of elements within data.

[0065] In certain implementations, the propagation field 108 may include a semantic field that identifies one or more semantic features that indicate contextual information regarding associated portions of the data 104. In certain implementations, the semantic features may include one or more categories, tags, contexts, classifications, or other indications that may be applied to data points, regions, or portions of the data 104. In certain implementations, classifications assigned by the semantic field may be predetermined (such as defined by one or more users, computing processes). In particular, the first machine learning model 110 and / or another machine learning model (such as the second machine learning model 112) may be trained to determine semantic masks with one or more predetermined categories. Example semantic masks include a background / foreground mask for image data, multi-class semantic segmentation masks, multi -identity instance segmentation masks, or a combination thereof. As one specific example, FIG. 5A depicts image data 500 along with exemplary semantic masks 502, 504, 506 according to one aspect of the present disclosure. The image data 500 depicts a first object (triangle), a second object (circle), and a third object (square). The semantic masks 502, 504, 506 may be determined to distinguish between different parts of the image. As depicted, the semantic mask 502 identifies portions of the image data 500 that correspond to the first object, the semantic mask 504 identifies portions of the image data that correspond to the second object, and the semantic mask 506 identifies portions of the image data 500 that correspond to the third object. In practice, semantic masks may combine multiple categories or classifications, and the same portion of data 104 may have multiple assigned categories. For example, a single semantic mask may be determined that combines all three of the semantic masks 502, 504, 506. As another example, the semantic mask 502 may be combined with a mask that identifies the foreground and background of the image. In such instances, the white region in the semantic mask 502 may be identified with multiple categories (such as “object 1” and “foreground”).

[0066] In certain implementations, the propagation field 108 may include an epistemic field that identifies predicted (epistemic) features within the data 104 based on previously- determined features within corresponding portions of previous data. In certain implementations, previous data may include previously received data within the data stream from which the data at 104 was received. For example, the computing device 102 may receive sequential sets, measurements, or frames of data and the data 104 may represent a current frame in processed within the data stream. In various implementations, the data stream may include a sequence of video image frames, a sequence of sensor measurements, and the like. In such implementations, the predicted features may be determined based on a predetermined number of previous frames of data received within the data stream (such as the previous 2 data frames, 5 data frames, 10 data frames, 100 data frames, and the like).

[0067] In one specific implementation, each feature in a dense feature map may be assumed to follow a particular arithmetic process (such as a Gaussian distribution / process). Under such an assumption, feature values for the features may be approximated as a Taylor series expansion based on previous observations (such as an V-th order Taylor expansion). For example, the predicted features within an epistemic field may be determined as:Where: ep iis the observation point (e.g., coordinates) of feature p in frame i, nFp(epi) is the feature value of feature p at the observation point ep i, Fp,i+i(x) is the predicted feature value of feature p at the point x.In certain such implementations, a machine learning model (such as the model 110 or another model 112) may be trained to determine the values of the Taylor expansion Fp,i+i( ) , such as based on a vector containing the previous feature values. In one particular implementation, a neural network may be trained to estimate a first or second order Taylor expansion for a given set of previous feature values (e.g., N=l, N=2). Insuch instances, the residual error may be negligible, while still allowing epistemic fields to be determined in a reasonable timeframe.

[0068] For example, FIG. 5B depicts image data 510 and a corresponding epistemic field 512 containing predicted features (such as predicted motion vectors) for corresponding pixels within the image data 510. In particular, the image data 510 may be received as part of a stream of image data captured from a front-facing camera positioned on a vehicle that is traveling forward. The epistemic field 512 may be determined based on previous motion vectors for previous image frames (such as the previous 5 image frames) captured by the camera. As can be seen in the epistemic field 512, based on the consistent forward direction of travel for the vehicle, the epistemic field 512 may have corresponding motion vector predictions based on the consistent motion in the preceding 5 image frames.

[0069] In certain implementations, the propagation field 108 may be determined at least in part based on extrinsic data that is received separately from the data 104. For example, the extrinsic data may include data that is not processed by the first machine learning model 110 during feature propagation. In certain implementations, the extrinsic data may include one or more propagation fields determined based on the data 104. For example, another computing device may determine and transmit the propagation field along with the data 104 (such as an initial computing device that receives and processes the data 104 before forwarding the data 104 to the computing device 102). For example, one or more of the masks 502, 504, 506 may be determined by another computing device and subsequently provided to the computing device 102.

[0070] In certain implementations, the extrinsic data may include one or more additional data sources that may be used when determining the propagation field 108. For example, the extrinsic data may include one or more data sources that may be used to construct a selection mask 106 or other attention mechanism that identifies portions of the data 104 for which a propagation field 108 is to be determined, as discussed further below. In still further implementations, data values from the extrinsic data may be used when determining the propagation field 108, but may not be used subsequently by the machine learning model 110 (such as during a feature propagation process 114).

[0071] In still further implementations, the extrinsic data may be used to select between different propagation fields determined by, or that may be determined by, the computing device 102. For example, the computing device 102 may be configured to determine one or more different types of propagation fields, including one or more different types of semantic fields, one or more different types of epistemic fields, one or more other types ofpropagation fields 108, or combinations thereof. As a further example, the computing device 102 may receive one or more different types of propagation fields in addition to or alternatively to determining different types of propagation fields on its own. In such implementations, the extrinsic data received from the computing device 102 may be used to select between the multiple types of propagation fields. In still further implementations, the computing device 102 may be configured to determine a propagation field 108 that focuses on or otherwise identifies particular portions of the data 104. For example, the computing device 102 may be configured to determine semantic fields that identify particular features or categories. In such instances, the extrinsic data may be used to indicate which types of features or which portions of the data 104 the propagation field 108 should identify.

[0072] In certain implementations, the extrinsic data may include user input, such as one or more prompts or requests received from a user (such as spoken prompts, text prompts, and the like). For example, the computing device 102 may include a display configured to display a user interface, and may receive user input via the user interface. In such instances, the user input may be used to select the type of propagation field 108 that is determined. As one specific example, the user input may request that the propagation field 108 focus on a particular object or type of object within image data (such as “focus on vehicles within the image”). In such instances, the computing device 102 may determine a propagation field 108 (such as a semantic field) that distinguishes between the specified object and other portions of received image data 104, such as the propagation field 524 in FIG. 5C. As another specific example, the user input may request that the propagation field 108 focus on a particular portion of the data (such as “focus on the center of the image”, “focus on sensor data received from ZIP Code 12345”, and the like). In such instances, the propagation field 108 may be determined to focus on the specified portions of the data (such as a semantic field that identifies the requested portions, a mask that identifies the requested portions, or combinations thereof).

[0073] In certain implementations, the propagation field 108 may be determined by the first machine learning model 110. For example, the propagation field 108 may be determined by the first machine learning model 110 before completing a feature propagation process 114. In additional or alternative implementations, the propagation field 108 may be determined by a second machine learning model 112 separate from the first machine learning model 110. For example, the second machine learning model 112 may determinethe propagation field 108 and may provide the propagation field 108 to the first machine learning model 110 for use while performing the feature propagation process 114.

[0074] In certain implementations, the propagation field 108 may be defined as a sparse field, a dense field, or a combination thereof. In certain such implementations, the propagation field 108 may be determined to have the same or similar density to a feature map determined by the first machine learning model 110 (such as to have the same density as output data 120 determined by the machine learning model 110). As one specific example, if the machine learning model 110 is configured to determine a sparse feature map, the propagation field 108 may similarly be determined with a sparse field. As another example, if the machine learning model 110 is configured to determine a dense feature map, the propagation field 108 may be similarly determined as a dense field. Sparse feature maps may represent maps where the majority of elements are zeroes or hold minimal significance, potentially indicating areas of low activity or importance within the received data. Conversely, dense feature maps may be characterized by an abundance of non-zero elements, suggesting a higher information content and a greater density of received data. For example, in a received image frame, a sparse feature map might emerge from an image with a basic background, where only selective features such as edges or specific objects are accentuated. On the other hand, a dense feature map might result from a complex scene where nearly every pixel may carry critical information, such as a scene rich in textures or with numerous interlaced objects, possibly necessitating a more comprehensive representation to encapsulate all the relevant features. Determined propagation fields 108 may be similarly dense or sparse.

[0075] In certain implementations, the computing device 102 may be further configured to determine a selection mask 106 for the data 104. The selection mask 106 may identify corresponding portions of the data 104 for which a propagation field 108 is to be determined. In certain implementations, the machine learning model 110 may be configured to perform the feature propagation process 114 within regions identified by the selection mask 106. As one specific example, it may be desirable to perform feature propagation only within a foreground of received image data. In such instances, the selection mask 106 may be determined to identify pixel locations within the image data that correspond to the foreground of the image. The computing device 102 may then determine the propagation field (such as according to one or more of the above-discussed techniques) within regions identified by the selection mask 106. In this way, the propagation field 108 may be limited to foreground portions of the received image data.In certain implementations, the portions of the data 104 identified by the selection mask 106 may be identified based on user input (such as a user prompt or request identifying a particular portion of the data). In such instances, the selection mask 106 may be determined according to the user input.

[0076] The computing device 102 may be configured to determine, with a first machine learning model 110, output data 120 based on the propagation field 108 and the data 104. In particular, the first machine learning model 110 may be configured to perform at least one feature propagation process 114 on the data 104 based on the propagation field 108 (such as based on latent features identified by the propagation field 108). For instance, the propagation field 108 may be used to guide the feature propagation process 114.

[0077] In one example implementation, performing the at least one feature propagation process 114 based on the propagation field 108 may include determining initial propagation values based on values of the propagation field 108. For example, the propagation field 108 may include and at the stomach field containing feature values predicted based on feature values for previous data frames. In such instances, the initialization step of the feature propagation process 114 may be performed using initialization values 116 based on the predicted feature values (such as equivalent to the predicted feature values). For example, after determining the initialization values 116 based on values within the propagation field, the computing device 102 (such as the machine learning model 110) may proceed with performing a propagation step of the feature propagation process 114 based on the initialization values 116.

[0078] In another example implementation, performing the at least one feature propagation process 114 based on the propagation field 108 may include comparing portions of the propagation field 108 to at least one corresponding threshold 118 and performing feature propagation for portions of the data 104 associated with portions of the propagation field 108 that satisfy the at least one corresponding threshold 118. In particular, the machine learning model 110 may be configured to exclude portions of the data 104 that do not satisfy the threshold 118. In certain implementations, portions of the data 104 may include individual data points (such as individual pixels, individual sensor measurements, and the like), or combinations thereof (such as clusters / combinations of two or more pixels, sensor measurements, and the like). In certain implementations, comparing portions of the propagation field 108 to the at least one corresponding threshold 118 may include comparing, to the at least one threshold 118, an average of values within the portions of the propagation field 108 to the threshold 118. In additional or alternativeimplementations, comparing portions of the propagation field 108 to the at least one corresponding threshold 118 may include comparing maximum values within the portions of the propagation field 108. In still further implementations, comparing portions of the propagation field 108 to the at least one corresponding threshold 118 may include comparing minimum values within portions of the propagation field 108. In further implementations, other values may be computed based on the values within corresponding portions of the propagation field 108 (such as weighted combinations, other statistical measures, and the like) and may be compared to the threshold 118. Certain implementations may combine one or more of the comparisons discussed above. In certain implementations, comparisons to the threshold 118 may serve as a stop condition for data frames. For example, if sufficient portions (such as a minimum number of portions or minimum percentage of the field 108) do not satisfy the threshold 118, feature propagation may not be performed for the data frame. In certain implementations, the data 104 may be multidimensional data 104. For example, the data 104 may be position data with x, y, and z coordinates and values. As another example, the data 104 may be image data with red, green, and blue values for each pixel. In instances where the data 104 is multidimensional, the at least one feature propagation process 114 may be performed on a per-dimension basis. Accordingly, comparing the data 104 to the threshold 118 may similarly be performed separately for data values in each dimension. The threshold 118 may specify one or more scalar quantities or vectors that determine whether certain actions should be taken, such as whether to include or exclude specific features in the propagation, or apply alternative propagation behaviors. For example, the threshold 118 may specify a confidence value for a predicted semantic feature that must be satisfied for the corresponding portion of the data to be included in the feature propagation process 114.

[0079] As a further example, performing the at least one feature propagation process 114 based on the propagation field 108 may include restricting feature propagation to only occur within portions of the data 104 having the same semantic category in the propagation field 108. In certain implementations, one or more of a propagation and a search step of the feature propagation process 114 may be limited to within portions of the data 104 that have the same value within one or more fields of the propagation field 108. For example, where the propagation field 108 includes one or more semantic fields that identify one or more categories or other semantic features for corresponding portions of the data 104, feature propagation may be limited to only occur within portions of the data that have thesame feature indicated within the semantic field. As a specific example, the semantic field may identify the foreground and the background of a received image data. In such instances, features from foreground pixels may only be propagated to other foreground pixels and features from back on pixels may only be propagated to other background pixels.

[0080] In certain implementations, the first machine learning model 110 may be updated over time. For example, the computing device 102 may be configured to receive at least a portion of the first machine learning model 110 (such as via a modem or other communication hardware). The computing device 102 may receive an initial version of the first machine learning model, updates to the first machine learning model 110, or combinations thereof. In such instances, the first machine learning model 110 may be trained by another computing device(s), such as one or more server devices.

[0081] In certain implementations, the computing device 102 may be configured to determine updates to the machine learning model 110 itself. For example, the computing device 102 may be configured to perform on-device training based on historical data received and processed by the first machine learning model (such as in addition to or alternatively to updates received from other computing devices. Such implementations may be beneficial for smaller models (such as 1.5 million parameters or less, 3 million parameters or less, and the like). In additional or alternative implementations, a portion or a submodule of the first machine learning model 110 could have adjustable weights through on-device training (such as for larger models), while keeping the rest of the model unchanged, which may enable on-device training for larger models.

[0082] In one particular implementation, the computing device 102 may be further configured to train the first machine learning model 110 to determine the propagation field 108. In particular, the training may be performed based on training data that specifies one or more expected outputs for received input data. During training, the model 110 may be configured to determine outputs based on provided input data and the resulting outputs may be compared to the expected outputs from the training data. In particular, crossentropy losses may be determined between the outputs produced by the model 110 and the expected outputs. Gradient backpropagation may then be performed based on the cross-entropy losses to updated one or more parameters, weights, or features used by the machine learning model, including one or more parameters, weights, or features used to determine the propagation field 108.

[0083] In the implementations discussed above, the computing device 102 is described as a determining and using a single propagation field 108. In additional or alternative implementations, it should be understood that more than one propagation field 108 may be used. For example, the machine learning model 110 may be configured to receive and utilize more than one propagation field 108 while performing the feature propagation process 114. To do so, the computing device 102 and / or the machine learning model 110 may be configured to combine the propagation fields while performing the feature propagation process 114. As one specific example, the machine learning model 110 may be configured to determine a weighted combination of two or more propagation fields and to use the weighted combination of the propagation fields to guide the feature propagation process 114 (such as using one or more of the guiding techniques discussed above). In implementations were more than one propagation field 108 is determined, the propagation fields may be determined by or received from different sources. For example, the machine learning model 110 may be configured to determine a first propagation field 108, receive a second propagation field from a second machine learning model 112, receive a third propagation fields from an extrinsic source (such as extrinsic data), and to determine a weighted combination of the first, second, and third propagation fields prior to performing the feature propagation process 114. Additional or alternative implementations utilizing more than one propagation field may be readily apparent based on the present disclosure to one skilled in the art. All such implementations are accordingly considered within the scope of the present disclosure.

[0084] In the implementations discussed above, the computing device 102 is described as performing various types of guidance of the feature propagation process 114 based on the propagation field 108. It should be appreciated that the provided implementations are merely exemplary and that, based on the accompanying disclosure, one skilled in the art will readily recognize various other techniques that may be used to guide a feature propagation process 114 based on a propagation field 108. All such implementations are considered within the scope of the present disclosure.

[0085] The examples discussed herein include one or more thresholds (such as the threshold 118) for making one or more determinations. In these examples, it should be understood that the provided conditions (e.g., greater than, less than, equal to, greater than or equal to, less than or equal to) are merely exemplary. Unless otherwise specified, alternative implementations of the described techniques may use different conditions. For example, a comparison that is described as using a “greater than” condition may be implementedwith a “less than or equal to" condition. All such implementations are considered within the scope of the present disclosure. Furthermmore, the phrase “satisfying a threshold” or similar may include a “greater than” condition, a “less than condition”, an “equal to condition”, or combinations thereof, depending on the implementation of the threshold, the comparison, or a combination thereof.

[0086] FIG. 4 is a flow chart illustrating an example method 400 for guiding feature propagation using one or more propagation fields according to one or more aspects of the present disclosure. The method may be performed by one or more of the above systems, such as the systems 100, 200, 300.

[0087] The method 400 includes receiving data for use by a first machine learning model (block 402). For example, the computing device 102 may receive data 104 for use by a first machine learning model 110.

[0088] The method 400 includes determining a propagation field for the data (block 404). For example, the computing device 102 may determine a propagation field 108 for the data 104. The propagation field 108 may identify latent features within the data 104. For example, the propagation field 108 may include a semantic field that identifies one or more semantic features that indicate contextual information regarding associated portions of the data 104. As another example, the propagation field 108 may include an epistemic field that identifies predicted features within the data 104 based on previously-predicted features within corresponding portions of previous data 104. In certain implementations, the propagation field 108 may be determined at least in part based on extrinsic data 104 that may be received separately from the data 104. For example, the extrinsic data 104 may be not processed during feature propagation to determine the output data 104. In certain such implementations, the determining the propagation field 108 may include determining, based on user input, a first type of propagation field 108 from among a plurality of types of propagation field 108 and determining the propagation field 108 based on the first type of propagation field 108. In certain implementations, the propagation field 108 may be determined by the first machine learning model 110. In certain implementations, the propagation field 108 may be determined by a second machine learning model 112 separate from the first machine learning model 110. In certain implementations, the propagation field 108 may be defined as a sparse field, a dense field, or a combination thereof, such as depending on the density of a corresponding feature map that the first machine learning model 110 is configured to determine. In certain implementations, prior to determining the propagation field 108, the method 400may include determining a selection mask 106 for the data 104. In such instances, the propagation field 108 may be determined within regions identified by the selection mask 106 and the feature propagation process 114 may be performed within regions identified by the selection mask 106.

[0089] The method 400 includes determining, with the first machine learning model, output data based on the propagation field and the data (block 406). For example, the computing device 102 may determine, with a first machine learning model 110, output data 120 based on the propagation field 108 and the data 104. While doing so, the first machine learning model 110 may perform at least one feature propagation process 114 based on the latent features identified by the propagation field 108. For example, performing the at least one feature propagation process 114 may include determining initial propagation values of the feature propagation process based on values of the propagation field 108. As another example, performing the at least one feature propagation process 114 may include comparing portions of the propagation field 108 to at least one corresponding threshold 118, and performing feature propagation for portions of the data 104 associated with portions of the propagation field 108 that satisfy the at least one corresponding threshold 118. In certain implementations, comparing portions of the propagation field 108 to the at least one corresponding threshold 118 may include comparing, to the at least one threshold 118, an average of values within the portions of the propagation field 108, maximum values within the portions of the propagation field 108, minimum values within portions of the propagation field 108, or a combination thereof. In certain implementations, the data 104 may be multidimensional data 104. In such instances, the at least one feature propagation process 114 may be performed on a per-dimension basis. In certain implementations, performing the at least one feature propagation process 114 comprises restricting feature propagation to only occur within portions of the data 104 having the same semantic category in the propagation field 108 (such as where the propagation field 108 includes a semantic field.

[0090] In certain implementations, the method 400 may further include training the first machine learning model 110 to determine the propagation field 108 using gradient backpropagation of cross-entropy losses against training data 104 for the output of the first machine learning model 110.

[0091] It is noted that one or more blocks (or operations) described with reference to FIG. 4 may be combined with one or more blocks (or operations) described with reference to anotherof the figures. For example, one or more blocks (or operations) of FIG. 4 may be combined with one or more blocks (or operations) of FIG. 1-3.

[0092] In one or more aspects, techniques for supporting vehicular operations may include additional aspects, such as any single aspect or any combination of aspects described below or in connection with one or more other processes or devices described elsewhere herein.

[0093] A first aspect provides a device that comprises one or more memories configured to receive and store input data for use by a first machine learning model; and one or more processors, coupled to the memories, configured to obtain the input data from the one or more memories, determine a propagation field for the input data, wherein the field identifies latent features within the input data, and determine, with the first machine learning model, output data based on the field and the input data. While determining the output data, the one or more processors are further configured to perform, with the first machine learning model, one or more feature propagation processes based on the latent features identified by the propagation field.

[0094] In a second aspect, in combination with the first aspect, the device further comprises a sensor configured to capture the input data.

[0095] In a third aspect, in combination with the second aspect, the sensor is a camera, and the input data comprises pixel data.

[0096] In a fourth aspect, in combination with at least one of the second aspect and the third, the sensor is a position sensor, and the input data comprises position data for one or more objects.

[0097] In a fifth aspect, in combination with at least one of the first through fourth aspects, the device further comprises a modem configured to receive at least a portion of the first machine learning model.

[0098] In a sixth aspect, in combination with at least one of the first through fifth aspects, performing the one or more feature propagation processes comprises determining initial propagation values based on values of the propagation field.

[0099] In a seventh aspect, in combination with at least one of the first through sixth aspects, the one or more processors are further configured, while performing the one or more feature propagation processes, to: compare portions of the propagation field to at least one corresponding threshold; and perform feature propagation for portions of the data associated only with portions of the propagation field that satisfy the at least one corresponding threshold.

[0100] In an eighth aspect, in combination with the seventh aspect, the one or more processors are further configured, while comparing portions of the propagation field to the at least one corresponding threshold, to: compare, to the at least one corresponding threshold, an average of values within the portions of the propagation field, maximum values within the portions of the propagation field, minimum values within portions of the propagation field, or a combination thereof.

[0101] In a ninth aspect, in combination with at least one of the first through eighth aspects, the data is multidimensional data, and wherein the one or more processors are configured to perform the one or more feature propagation processes on a per-dimension basis.

[0102] In a tenth aspect, in combination with at least one of the first through ninth aspects, the propagation field includes a semantic field that identifies one or more semantic features that indicate contextual information regarding associated portions of the data.

[0103] In an eleventh aspect, in combination with the tenth aspect, the one or more processors are further configured, while performing the one or more feature propagation processes, to: restrict feature propagation to only occur within portions of the data having the same semantic category in the propagation field.

[0104] In a twelfth aspect, in combination with at least one of the first through eleventh aspects, the field includes an epistemic field that identifies predicted features within the data based on previously-predicted features within corresponding portions of previous data.

[0105] In a thirteenth aspect, in combination with at least one of the first through twelfth aspects, the field is determined at least in part based on extrinsic data that is received separately from the data.

[0106] In a fourteenth aspect, in combination with at least one of the first through thirteenth aspects, the device further comprises a display device, configured to display a user interface, and wherein the one or more processors are further configured to receive user input via the user interface.

[0107] In a fifteenth aspect, in combination with the fourteenth aspect, the one or more processors are further configured to: determine, based on user input, a first type of propagation field from among a plurality of types of propagation field; and determine the propagation field based on the first type of propagation field.

[0108] In a sixteenth aspect, in combination with at least one of the first through fifteenth aspects, the propagation field is determined by the first machine learning model.

[0109] In a seventeenth aspect, in combination with the sixteenth aspect, wherein the one or more processors are further configured to train the first machine learning model to determinethe propagation field using gradient backpropagation of cross-entropy losses against training data for the output of the first machine learning model.

[0110] In an eighteenth aspect, in combination with at least one of the first through seventeenth aspects, the one or more processors are further configured to determine the propagation field using a second machine learning model separate from the first machine learning model.[OHl] In a nineteenth aspect, in combination with at least one of the first through eighteenth aspects, the one or more processors are further configured to: determine a selection mask for the data, wherein the propagation field is determined within regions identified by the selection mask, and wherein the feature propagation process is performed within regions identified by the selection mask.

[0112] In a twentieth aspect, in combination with one or more of the first aspect through the nineteenth aspect, the propagation field is defined as a sparse field, a dense field, or a combination thereof.

[0113] A twenty-first aspect provides a method that includes receiving data for use by a first machine learning model; determining a propagation field for the data, wherein the field identifies latent features within the data; and determining, with the first machine learning model, output data based on the field and the data, wherein the first machine learning model performs at least one feature propagation process based on the latent features identified by the propagation field.

[0114] A twenty-second aspect includes a non-transitory, computer-readable medium storing instructions which, when executed by a processor, cause the processor to: receive data for use by a first machine learning model; determine a propagation field for the data, wherein the field identifies latent features within the data; and determine, with the first machine learning model, output data based on the field and the data, wherein the first machine learning model performs at least one feature propagation process based on the latent features identified by the propagation field.

[0115] In some implementations, the system includes a wireless device, such as a UE. In some implementations, the system may include at least one processor, and a memory coupled to the processor. The processor may be configured to perform operations described herein with respect to the apparatus. In some other implementations, the system may include a non-transitory computer-readable medium having program code recorded thereon and the program code may be executable by a computer for causing the computer to perform operations described herein with reference to the system. In some implementations, thesystem may include one or more means configured to perform operations described herein. In some implementations, a method of wireless communication may include one or more operations described herein with reference to the system.

[0116] Components, the functional blocks, and the modules described herein with respect to FIGs. 1-4 include processors, electronics devices, hardware devices, electronics components, logical circuits, memories, software codes, firmware codes, among other examples, or any combination thereof. Software shall be construed broadly to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, application, software applications, software packages, routines, subroutines, objects, executables, threads of execution, procedures, and / or functions, among other examples, whether referred to as software, firmware, middleware, microcode, hardware description language or otherwise. In addition, features discussed herein may be implemented via specialized processor circuitry, via executable instructions, or combinations thereof.

[0117] Those of skill would further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the disclosure herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure. Skilled artisans will also readily recognize that the order or combination of components, methods, or interactions that are described herein are merely examples and that the components, methods, or interactions of the various aspects of the present disclosure may be combined or performed in ways other than those illustrated and described herein.

[0118] The various illustrative logics, logical blocks, modules, circuits and algorithm processes described in connection with the implementations disclosed herein may be implemented as electronic hardware, computer software, or combinations of both. The interchangeability of hardware and software has been described generally, in terms of functionality, and illustrated in the various illustrative components, blocks, modules, circuits and processes described above. Whether such functionality is implemented inhardware or software depends upon the particular application and design constraints imposed on the overall system.

[0119] The hardware and data processing apparatus used to implement the various illustrative logics, logical blocks, modules and circuits described in connection with the aspects disclosed herein may be implemented or performed with a general purpose single- or multi-chip processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, or, any conventional processor, controller, microcontroller, or state machine. In some implementations, a processor may be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. In some implementations, particular processes and methods may be performed by circuitry that is specific to a given function.

[0120] In one or more aspects, the functions described may be implemented in hardware, digital electronic circuitry, computer software, firmware, including the structures disclosed in this specification and their structural equivalents thereof, or in any combination thereof. Implementations of the subject matter described in this specification also may be implemented as one or more computer programs, that is one or more modules of computer program instructions, encoded on a computer storage media for execution by, or to control the operation of, data processing apparatus.

[0121] If implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. The processes of a method or algorithm disclosed herein may be implemented in a processor-executable software module which may reside on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that may be enabled to transfer a computer program from one place to another. A storage media may be any available media that may be accessed by a computer. By way of example, and not limitation, such computer-readable media may include random-access memory (RAM), read-only memory (ROM), electrically erasable programmable readonly memory (EEPROM), CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that may be used to store desiredprogram code in the form of instructions or data structures and that may be accessed by a computer. Also, any connection may be properly termed a computer-readable medium. Disk and disc, as used herein, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk, and Blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media. Additionally, the operations of a method or algorithm may reside as one or any combination or set of codes and instructions on a machine readable medium and computer-readable medium, which may be incorporated into a computer program product.

[0122] Various modifications to the implementations described in this disclosure may be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to some other implementations without departing from the spirit or scope of this disclosure. Thus, the claims are not intended to be limited to the implementations shown herein, but are to be accorded the widest scope consistent with this disclosure, the principles and the novel features disclosed herein.

[0123] Certain features that are described in this specification in the context of separate implementations also may be implemented in combination in a single implementation. Conversely, various features that are described in the context of a single implementation also may be implemented in multiple implementations separately or in any suitable subcombination. Moreover, although features may be described above as acting in certain combinations and even initially claimed as such, one or more features from a claimed combination may in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination.

[0124] Similarly, while operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. Further, the drawings may schematically depict one more example processes in the form of a flow diagram. However, other operations that are not depicted may be incorporated in the example processes that are schematically illustrated. For example, one or more additional operations may be performed before, after, simultaneously, or between any of the illustrated operations. In certain circumstances, multitasking and parallel processing may be advantageous. Moreover, the separation of various system components in the implementations described above should not be understood as requiring such separation in all implementations, and it should beunderstood that the described program components and systems may generally be integrated together in a single software product or packaged into multiple software products. Additionally, some other implementations are within the scope of the following claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve desirable results.

[0125] The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

CLAIMSWHAT IS CLAIMED IS:

1. A device comprising: one or more memories configured to receive and store input data for use by a first machine learning model; one or more processors, coupled to the memories, configured to: obtain the input data from the one or more memories; determine a propagation field for the input data, wherein the field identifies latent features within the input data; and determine, with the first machine learning model, output data based on the field and the input data, wherein, while determining the output data, the one or more processors are configured to: perform, with the first machine learning model, one or more feature propagation processes based on the latent features identified by the propagation field.

2. The device of claim 1, further comprising a sensor configured to capture the input data.

3. The device of claim 2, wherein the sensor is a camera, and the input data comprises pixel data.

4. The device of claim 2, wherein the sensor is a position sensor, and the input data comprises position data for one or more objects.

5. The device of claim 1, further comprising a modem configured to receive at least a portion of the first machine learning model.

6. The device of claim 1, wherein performing the one or more feature propagation processes comprises determining initial propagation values based on values of the propagation field.

7. The device of claim 1, wherein the one or more processors are further configured, while performing the one or more feature propagation processes, to: compare portions of the propagation field to at least one corresponding threshold; and perform feature propagation for portions of the data associated only with portions of the propagation field that satisfy the at least one corresponding threshold.

8. The device of claim 7, wherein the one or more processors are further configured, while comparing portions of the propagation field to the at least one corresponding threshold, to: compare, to the at least one corresponding threshold, an average of values within the portions of the propagation field, maximum values within the portions of the propagation field, minimum values within portions of the propagation field, or a combination thereof.

9. The device of claim 7, wherein the data is multidimensional data, and wherein the one or more processors are configured to perform the one or more feature propagation processes on a per-dimension basis.

10. The device of claim 1, wherein the propagation field includes a semantic field that identifies one or more semantic features that indicate contextual information regarding associated portions of the data.

11. The device of claim 10, wherein the one or more processors are further configured, while performing the one or more feature propagation processes, to: restrict feature propagation to only occur within portions of the data having the same semantic category in the propagation field.

12. The device of claim 1, wherein the field includes an epistemic field that identifies predicted features within the data based on previously -predicted features within corresponding portions of previous data.

13. The device of claim 1, wherein the field is determined at least in part based on extrinsic data that is received separately from the data.

14. The device of claim 1, further comprising a display device, configured to display a user interface, and wherein the one or more processors are further configured to receive user input via the user interface.

15. The device of claim 14, wherein the one or more processors are further configured to: determine, based on the user input, a first type of propagation field from among a plurality of types of propagation field; anddetermine the propagation field based on the first type of propagation field.

16. The device of claim 1, wherein the propagation field is determined by the first machine learning model.

17. The device of claim 16, wherein the one or more processors are further configured to train the first machine learning model to determine the propagation field using gradient backpropagation of cross-entropy losses against training data for the output of the first machine learning model.

18. The device of claim 1, wherein the one or more processors are further configured to determine the propagation field using a second machine learning model separate from the first machine learning model.

19. The device of claim 1, wherein the one or more processors are further configured to: determine a selection mask for the data, wherein the propagation field is determined within regions identified by the selection mask, and wherein the feature propagation process is performed within regions identified by the selection mask.

20. The device of claim 1, wherein the propagation field is defined as a sparse field, a dense field, or a combination thereof.

21. A method compri sing : receiving data for use by a first machine learning model; determining a propagation field for the data, wherein the field identifies latent features within the data; and determining, with the first machine learning model, output data based on the field and the data, wherein the first machine learning model performs at least one feature propagation process based on the latent features identified by the propagation field.

22. A non-transitory, computer-readable medium storing instructions which, when executed by a processor, cause the processor to: receive data for use by a first machine learning model; determining a propagation field for the data, wherein the field identifies latent features within the data; and determining, with the first machine learning model, output data based on the field and the data, wherein the first machine learning model performs at least one feature propagation process based on the latent features identified by the propagation field.

Images