Information processing system, information processing method and program

Abstract

The present invention suppresses deterioration in the accuracy of subsequent processing due to differences in image capture conditions.　This information processing system: acquires a first image of a subject captured under first image capture conditions; acquires a second image of the subject captured under second image capture conditions; extracts a first feature quantity indicating a feature of the first image, and a second feature quantity indicating a feature of a transformed second image produced by an image transformation unit, which is a machine learning model that receives input of the acquired original second image and outputs the transformed second image; and trains the image transformation unit on the basis of the difference in characteristics between the first and second feature quantities for a plurality of subjects.

Description

[Technical Field]

[0001] The present invention relates to an information processing system, an information processing method, and a program. [Background technology]

[0002] In recent years, image processing using machine learning models has become widespread. For example, a machine learning model receives input information representing a captured image, and outputs some kind of result. Summary of the Invention [Problem to be solved by the invention]

[0003] For example, if the camera used to capture images used to train a machine learning model is different from the camera used to capture images to input into the trained machine learning model, the characteristics of the images may differ due to different shooting conditions during training and inference. In such cases, there is a risk that the accuracy of the machine learning model's output may decrease.

[0004] The present invention has been made in consideration of the above-mentioned problems, and its purpose is to provide a technology that suppresses a decrease in the accuracy of processing when an image captured under certain shooting conditions and an image captured under other shooting conditions are processed using the same method, for example, including the same machine learning model. [Means for solving the problem]

[0005] In order to solve the above problem, the information processing system of the present invention includes a first image acquisition unit that acquires a first image of a subject captured under a first shooting condition; a second image acquisition unit that acquires a second image of the subject captured under a second shooting condition; an image conversion unit that is a machine learning model that receives the acquired second image as input and outputs a converted second image; an extraction unit that extracts a first feature amount that indicates a characteristic of the first image and a second feature amount that indicates a characteristic of the converted second image from the second image converted by the image conversion unit; and a conversion learning unit that trains the image conversion unit based on differences in characteristics between the first feature amount and the second feature amount for a plurality of subjects.

[0006] In one form of the present invention, the first shooting condition may be to shoot using a first camera, and the second shooting condition may be to shoot using a second camera having characteristics different from those of the first camera.

[0007] In one form of the present invention, the conversion learning unit may train the image conversion unit based on the difference between a value output when the first feature or the second feature is input and a correct answer, using a classifier trained with training data including a plurality of input data each including either the first feature or the second feature and correct answer data indicating that each of the input data is either the first feature or the second feature.

[0008] In one form of the present invention, the information processing system may further include a processing unit that inputs features extracted by the extraction unit from an image captured under the second shooting conditions and converted by the image conversion unit into a machine learning model trained based on features extracted by the extraction unit from an image captured under the first shooting conditions, and performs processing based on the output of the machine learning model for the input features.

[0009] In one aspect of the present invention, the machine learning model may output information indicating whether or not an image captured under the second shooting conditions and converted by the image conversion unit includes a predetermined object.

[0010] Furthermore, an information processing method according to the present invention includes the steps of acquiring a first image of a subject captured under a first shooting condition, acquiring a second image of the subject captured under a second shooting condition, extracting a first feature amount indicating a characteristic of the first image, and a second feature amount indicating a characteristic of the second image converted by an image conversion unit that is a machine learning model that receives the acquired second image as input and outputs a converted second image, and training the image conversion unit based on differences in characteristics between the first feature amount and the second feature amount for a plurality of subjects.

[0011] Furthermore, the program of the present invention causes a computer to execute a process of acquiring a first image of a subject taken under a first shooting condition, acquiring a second image of the subject taken under a second shooting condition, extracting a first feature amount indicating the characteristics of the first image and a second feature amount indicating the characteristics of the second image converted by an image conversion unit, which is a machine learning model that receives the acquired second image as input and outputs a converted second image, and training the image conversion unit based on differences in characteristics between the first feature amount and the second feature amount for multiple subjects. [Effects of the Invention]

[0012] According to the present invention, when an image taken under certain shooting conditions and an image taken under other shooting conditions are processed using the same method, for example, including the same machine learning model, it is possible to suppress a decrease in the accuracy of the processing. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is a diagram illustrating an example of a configuration of an information processing system according to an embodiment of the present invention. [Figure 2]FIG. 2 is a functional block diagram showing an example of functions implemented in an information processing system according to an embodiment of the present invention. [Figure 3] FIG. 2 is a functional block diagram showing an example of functions implemented in an information processing system according to an embodiment of the present invention. [Figure 4] FIG. 10 is a flow chart showing an outline of a process for training an image conversion unit. [Figure 5] FIG. 10 is a diagram showing an example of a captured image. [Figure 6] FIG. 10 is a flowchart illustrating an example of a process for training a camera classifier. [Figure 7] FIG. 10 is a flowchart showing an example of a process for training an image conversion unit. [Figure 8] FIG. 10 is a flowchart showing an example of a learning process for an object identification model based on a first camera. [Figure 9] FIG. 10 is a flow diagram illustrating an example of an inference process based on a second camera. DETAILED DESCRIPTION OF THE INVENTION

[0014] An embodiment of the present invention will now be described in detail with reference to the accompanying drawings. In this embodiment, the present invention is applied to an information processing system that receives an input image of a subject object and determines whether the image contains a target object.

[0015] This information processing system uses a machine learning model to determine whether an image contains a target object. Furthermore, the conditions under which an image to be determined during inference (e.g., the camera used for capturing the image) are taken are different from the conditions under which images included in the training data for the machine learning model are captured. Below, we describe an information processing system configured to accommodate such differences in capturing conditions, particularly differences in cameras.

[0016] FIG. 1 is a diagram showing an example of the configuration of an information processing system according to one embodiment of the present invention. The information processing system according to this embodiment includes an information processing device 10. The information processing device 10 is, for example, a computer such as a game console or a personal computer. As shown in FIG. 1, the information processing device 10 includes, for example, a processor 11, a storage unit 12, a communication unit 14, an operation unit 16, a display unit 18, a first camera 20a, and a second camera 20b. The information processing system may be configured by one information processing device 10, or may be configured by multiple devices including the information processing device 10.

[0017] The processor 11 is a program-controlled device such as a CPU that operates according to a program installed in the information processing device 10, for example.

[0018] The storage unit 12 is composed of at least a part of a storage element such as a ROM or a RAM, or an external storage device such as a solid state drive. The storage unit 12 stores programs executed by the processor 11, etc.

[0019] The communication unit 14 is a communication interface for wired or wireless communication, such as a network interface card, and transmits and receives data to and from other computers or terminals via a computer network such as the Internet.

[0020] The operation unit 16 is an input device such as a keyboard, a mouse, a touch panel, or a game console controller, and receives operation input from the user and outputs a signal indicating the content of the input to the processor 11.

[0021] The display unit 18 is a display device such as a liquid crystal display, and displays various images according to instructions from the processor 11. The display unit 18 may be a device that outputs a video signal to an external display device.

[0022] The first camera 20a and the second camera 20b are imaging devices including an imaging element, and are, for example, cameras capable of capturing moving images. The first camera 20a and the second camera 20b may be cameras capable of acquiring visible RGB images. The first camera 20a and the second camera 20b may be cameras capable of acquiring visible RGB images and depth information synchronized with the RGB images.

[0023] The first camera 20a and the second camera 20b have different characteristics. The characteristics include, for example, gamma characteristics, image distortion, F-number, focal length, and the presence or absence of an image stabilization function, and at least one of these characteristics is different between the first camera 20a and the second camera 20b. At least one of the first camera 20a and the second camera 20b may be external to the information processing device 10 or may be built into another computer. At least one of the first camera 20a and the second camera 20b may be connected to the information processing device 10 via the communication unit 14 or an input / output unit described below.

[0024] The information processing device 10 may include audio input / output devices such as a microphone, a speaker, etc. The information processing device 10 may also include, for example, a communication interface such as a network board, an optical disc drive that reads optical discs such as DVD-ROMs and Blu-ray (registered trademark) discs, and an input / output unit (a USB (Universal Serial Bus) port) for inputting and outputting data to and from external devices.

[0025] 2 and 3 are functional block diagrams showing an example of functions implemented in an information processing system according to an embodiment of the present invention. Fig. 2 shows a function for training the image conversion unit 22 to handle differences between cameras, and Fig. 3 shows a function for training an object identification model 24 using the trained image conversion unit 22 and identifying an image using the object identification model 24.

[0026] 2 and 3, the information processing system functionally includes a first image acquisition unit 21a, a second image acquisition unit 21b, an image conversion unit 22, a feature extraction unit 23, a first feature extraction unit 23a, a second feature extraction unit 23b, an object identification model 24, a conversion learning unit 31, a difference detection unit 32, and an object learning unit 35. The difference detection unit 32 includes a camera classifier 33, which is a type of machine learning model. The object identification model 24 is an example of a processing unit that executes processing according to the output of the second feature extraction unit 23b.

[0027] These functions are implemented mainly by the processor 11 and the storage unit 12. More specifically, these functions may be implemented by having the processor 11 execute a program that is installed in the information processing device 10, which is a computer, and that includes execution instructions corresponding to the above functions. Furthermore, this program may be supplied to the information processing device 10 via a computer-readable information storage medium such as an optical disk, a magnetic disk, or a flash memory, or via the Internet, for example.

[0028] Note that the information processing system according to this embodiment does not necessarily have to be equipped with all of the functions shown in FIG. 2, and may also be equipped with functions other than those shown in FIG.

[0029] The first image acquisition unit 21a acquires a first image captured by the first camera 20a. In this embodiment, the first image is mainly an image of at least one of a plurality of objects. Note that an object is an example of a subject, and the first image may also be an image of another type of subject, for example, a subject that does not have a clear shape. Unless otherwise specified below, an image of an object may also be an image of another type of subject.

[0030] The second image acquisition unit 21b acquires a second image captured by the second camera 20b. In this embodiment, the second image is mainly an image of the same object as the first image.

[0031] When a second image is input, the image conversion unit 22 outputs the converted second image. The image conversion unit 22 is a machine learning model. The learning method of the image conversion unit 22 will be described in detail later.

[0032] The feature extraction unit 23 extracts features indicating the characteristics of each image from multiple images including a first image and a converted second image. Hereinafter, a feature extracted from a first image will be referred to as a first feature, and a feature extracted from a second image will be referred to as a second feature. The first feature extraction unit 23a extracts a first feature indicating the characteristics of the first image from the first image. The second feature extraction unit 23b extracts a second feature indicating the characteristics of the converted second image from the second image converted by the image conversion unit 22. The feature extraction unit 23, the first feature extraction unit 23a, and the second feature extraction unit 23b may be the same entity or may be located on different computers. Even if located on different computers, the feature extraction unit 23, the first feature extraction unit 23a, and the second feature extraction unit 23b extract features using the same method.

[0033] The difference detection unit 32 detects differences in characteristics between the first feature amount and the second feature amount for a plurality of objects. The camera classifier 33 included in the difference detection unit 32 is trained using training data including a plurality of input data each including either the first feature amount or the second feature amount, and correct answer data indicating that each of the input data is either the first feature amount or the second feature amount. The difference detection unit 32 detects, as a difference in characteristics, a difference between a value output when the first feature amount or the second feature amount is input to the camera classifier 33 and the correct answer.

[0034] The conversion learning unit 31 causes the image conversion unit 22 to learn based on the result of the detection of the difference in characteristics by the difference detection unit 32.

[0035] The object identification model 24 is a type of machine learning model, such as an SVM (Support Vector Machine). The object identification model 24 is trained using features extracted by a first feature extraction unit 23a from an image captured by a first camera 20a. In this embodiment, the object identification model 24 outputs information indicating whether an image includes a target object, based on features extracted by a second feature extraction unit 23b from an image captured by a second camera 20b and converted by a trained image conversion unit 22.

[0036] More specifically, in response to input of data indicating feature quantities corresponding to an image, the object identification model 24 outputs an identification score indicating the probability that an object appearing in the image belongs to a positive class in the object identification model 24. The object identification model 24 is trained using a plurality of positive example training data for positive examples and a plurality of negative example training data for negative examples. The positive example training data is generated by the first feature extraction unit 23a from positive example images including images of a target object, and the negative example training data is generated by the first feature extraction unit 23a from negative example images of an object different from the target object. The positive example images may be images captured by the first camera 20a, and the negative example images may be images of the environment of the first camera 20a captured by the first camera 20a. Note that the positive example images may also be images captured by the second camera 20b and converted by the image conversion unit 22.

[0037] Here, the first feature extraction unit 23a includes a trained CNN (Convolutional Neural Network). In response to an input image, this CNN outputs data indicating feature amounts corresponding to the image. Metric learning has been performed in advance on the CNN of the first feature extraction unit 23a. Through the advance metric learning, the CNN is tuned so as to output feature amount data indicating feature amounts that are similar to each other for images that include objects belonging to the positive class in the object identification model 24. The feature amounts according to this embodiment are, for example, vector amounts normalized so that the norm is 1. Note that this metric learning may be performed before capturing an image of the target object.

[0038] By using a CNN on which metric learning has been performed in advance, the features of samples belonging to one class are aggregated into a compact region regardless of conditions. As a result, the information processing device 10 according to this embodiment is able to determine a valid discrimination boundary for the object identification model 24 even from a small number of samples. This allows the object identification model 24 to be trained in a short time. The expected time required is, for example, several tens of seconds to grasp and rotate the object to acquire a positive example image, and a few seconds for machine learning.

[0039] In this embodiment, it is desirable to perform metric learning on the CNN to ensure overall accuracy, but this is not essential. The first feature extraction unit 23a may output data indicating the feature amounts corresponding to an input image using another known algorithm that calculates the feature amounts indicating the characteristics of the image.

[0040] The object learning unit 35 trains the object identification model 24 using training data. The training data for the object identification model 24 includes a plurality of positive example feature amounts and a plurality of negative example feature amounts. The plurality of negative example feature amounts are extracted by the first feature extraction unit 23a from a plurality of negative example images that are captured by the first camera 20a and do not contain a target object. The plurality of positive example feature amounts are extracted by the first feature extraction unit 23a from a plurality of positive example images that include an image of a target object captured by the first camera 20a. The positive example feature amounts may be extracted by the first feature extraction unit 23a or the second feature extraction unit 23b from a positive example image obtained by converting an image of a target object captured by the second camera 20b by the image conversion unit 22.

[0041] Next, we will explain the learning process of the image conversion unit 22. In this embodiment, the image conversion unit 22 is trained by a generative adversarial network method using the camera classifier 33 as a discriminator. Figure 4 is a flow diagram that schematically shows the process of training the image conversion unit 22.

[0042] First, the conversion learning unit 31 stores a set of a first image captured by the first camera 20a and a second image captured by the second camera 20b in the memory unit 12 (S101). A plurality of sets are stored in the memory unit 12. Each set includes a first image and a second image. The first image and the second image included in the same set are images of the same object. The plurality of sets includes images of each of a plurality of objects. The conversion learning unit 31 acquires the first image from the first image acquisition unit 21a and acquires the second image from the second image acquisition unit 21b.

[0043] Fig. 5 is a diagram showing an example of a captured image. Fig. 5 shows an image of a tool as an example of an object. Fig. 5 is an example of a first image or a second image. The first image acquisition unit 21a may acquire the entire area of ​​the image captured by the first camera 20a as the first image, or may acquire only a part of the image captured by the first camera 20a as the first image.

[0044] In the latter case, the first image acquisition unit 21a may input an image captured by the first camera 20a into a pre-trained Regional Proposal Network (RPN) and acquire, as the first image, an area in which the RPN estimates that some object exists. Similarly, the second image acquisition unit 21b may acquire, as the second image, a part or all of the area in the image captured by the second camera 20b. The second image acquisition unit 21b may also acquire, as the second image, an area in which the RPN estimates that some object exists.

[0045] When the plurality of sets are stored in the storage unit 12, the conversion learning unit 31 trains the camera classifier 33 based on the first image and the second image included in the plurality of sets (S102). Meanwhile, the conversion learning unit 31 adjusts the parameters of the image conversion unit 22 based on the first image and the second image so that the camera classifier 33 does not classify the first image and the second image (S103).

[0046] If the predetermined learning termination condition is not met (N in S104), the conversion learning unit 31 repeats the processing from S102 onwards. If the predetermined learning termination condition is met (Y in S104), the conversion learning unit 31 terminates the processing. The learning termination condition may simply be the number of repetitions, or may be that a state in which the parameter change rate is smaller than the threshold has been detected a predetermined number of times recently.

[0047] The processing in S102 will be described in further detail. Fig. 6 is a flow diagram showing an example of processing for training the camera classifier 33. The conversion learning unit 31 trains the camera classifier 33 so that the camera classifier 33 can classify images captured by the first camera 20a. Note that information indicating whether an input image is the first image or the second image is not input to the camera classifier 33 itself.

[0048] First, the conversion learning unit 31 acquires training data including training images including a plurality of first images and a plurality of second images, and correct answer data (S201), from a plurality of sets stored in the storage unit 12. The correct answer data is data indicating whether the training image is a first image or a second image.

[0049] When training data including training images is acquired, the first feature extraction unit 23a or the second feature extraction unit 23b extracts features from each of the multiple training images (S202). The transformation learning unit 31 inputs each of the extracted features to the camera classifier 33 (S203). The transformation learning unit 31 adjusts the parameters of the camera classifier 33 for each training image based on the output of the camera classifier 33 and the ground truth data (S204).

[0050] The conversion learning unit 31 adjusts the parameters of the camera classifier 33 based on the output of the camera classifier 33 and the correct answer data (S204). In adjusting the parameters, the conversion learning unit 31 updates the parameters so as to minimize the binary cross entropy between the output of the camera classifier 33 and the true value. Note that the parameters may also be updated by other methods such as mean squared error.

[0051] After the parameters are adjusted, the conversion learning unit 31 determines whether a termination condition for learning of the camera classifier 33 is satisfied (S209). If the termination condition is not satisfied (N in S209), the processing from S202 onward is repeated. If the termination condition is satisfied (Y in S209), the processing in FIG. 6 is terminated. The termination condition may be that the number of times the processing is repeated reaches a predetermined number (number of epochs), or that the accuracy rate of the camera classifier 33 for test input data exceeds a threshold. The conversion learning unit 31 may store the feature values ​​extracted in S202 in the storage unit 12, and may acquire the feature values ​​stored in the storage unit 12 instead of the processing in S202 from the second time onward. In this case, only the processing in S203 and S204 may be repeated.

[0052] The process in S103 will be described in more detail below. Fig. 7 is a flow chart showing an example of a process for training the image conversion unit 22.

[0053] First, the conversion learning unit 31 causes the image conversion unit 22 to convert second images included in the multiple sets stored in the storage unit 12 (S251). Then, the conversion learning unit 31 acquires training images including multiple first images included in the multiple sets and multiple converted second images, and training data including correct answer data (S252). The correct answer data is data indicating whether the training image is a first image or a second image.

[0054] Next, the first feature extraction unit 23a or the second feature extraction unit 23b extracts features from each of the multiple training images (S253). The difference detection unit 32 acquires the output of the camera classifier 33 to which the extracted features have been input (S254). Note that hereinafter, the output of the camera classifier 33 for the first image as a training image will be referred to as the first output, and the output of the camera classifier 33 for the second image as a training image will be referred to as the second output.

[0055] Then, the transformation learning unit 31 adjusts the parameters of the image transformation unit 22 for each training image based on the output of the camera classifier 33 and the ground truth data (S255). In adjusting the parameters, the transformation learning unit 31 updates the parameters so that the binary cross entropy between the output of the camera classifier 33 and the true value increases. Note that the parameters may also be updated using other methods such as mean squared error. Note that, although the parameter adjustment is performed collectively for the first image and the second image in the example of FIG. 7, it may also be performed at separate times.

[0056] Once the parameters are adjusted, the conversion learning unit 31 determines whether the termination condition for learning of the image conversion unit 22 is met (S259). If the termination condition is not met (N in S259), the processing from S251 onwards is repeated, and if the termination condition is met (Y in S259), the processing in Fig. 7 ends. The termination condition may be that the number of times the processing is repeated reaches a predetermined number (number of epochs), or that the accuracy rate of the camera classifier 33 for test input data becomes lower than a threshold.

[0057] In this embodiment, in a system premised on feature extraction, the image conversion unit 22 is trained not by distinguishing between a first image and a converted second image, but by distinguishing between features extracted from a first image and features extracted from a converted second image. When processing is performed on images captured under certain shooting conditions and images captured under other shooting conditions, the image conversion unit 22 can more appropriately convert the images and suppress a decrease in the accuracy of the processing. Furthermore, because the image conversion unit 22 can perform relatively weak transformations required to extract features, training is easier than matching the images themselves, and problems such as distortion that accompany strong transformations can also be suppressed.

[0058] Furthermore, since information is abstracted by extracting features, it becomes possible to train the image conversion unit 22 appropriately even when, for example, the first camera 20a and the second camera 20b photograph an object from different positions and / or angles.

[0059] The difference detection unit 32 determines the accuracy of the output of the camera classifier 33 each time the first image or the transformed second image is input to the camera classifier 33. Meanwhile, the camera classifier 33 is trained to distinguish between the first image and the transformed second image, and the classification is performed internally by focusing on the different characteristics between the first feature amount and the second feature amount. Therefore, the camera classifier 33's correct classification indicates that it has properly detected the difference in characteristics between the first feature amount and the second feature amount. Therefore, the determination of the accuracy of the output of the camera classifier 33 by the difference detection unit 32 corresponds to the detection of the difference in characteristics between the first feature amount and the second feature amount.

[0060] Next, an example of use of the trained image conversion unit 22 will be described. Fig. 8 is a flow diagram showing an example of a training process for the object identification model 24 based on the first camera 20a. The process shown in Fig. 8 is an example of a process using the configuration shown on the left side of Fig. 3, and is an example of a case where an image for training the object identification model 24 is captured by the first camera 20a.

[0061] First, the first image acquisition unit 21a acquires a plurality of positive example images in which the target object is photographed by the first camera 20a (S301). The first image acquisition unit 21a acquires a plurality of negative example images that are photographed by the first camera 20a and do not include the target object (S302).

[0062] The first feature extraction unit 23a extracts features (positive example features) from each positive example image (S303), and extracts features (negative example features) from each negative example image (S304). The object learning unit 35 then trains the object identification model 24 based on the positive example features and the negative example features (S305). The object learning unit 35 sets the trained object identification model 24 to be used for inference (S306). More specifically, the setting may involve copying parameters of the trained object identification model 24 to parameters of the object identification model 24 for inference, or may involve setting the trained object identification model 24 itself so that features to be inferred are input.

[0063] The positive example images may not be images captured by the first camera 20a, but may be images captured by the second camera 20b and converted by the image conversion unit 22. The negative example images may be a mixture of images captured by the first camera 20a and images captured by the second camera 20b and converted by the image conversion unit 22. Even in such cases, the object identification model 24 can be properly trained.

[0064] Fig. 9 is a flow diagram showing an example of inference processing based on the second camera 20b. Fig. 9 shows an example of processing for determining whether an image captured by the second camera 20b includes a target object using the trained object identification model 24. Fig. 9 also describes processing when it is determined that the image includes a target object.

[0065] First, the second image acquisition unit 21b acquires an image captured by the second camera 20b (S351). The trained image conversion unit 22 converts the acquired image (S352). The second feature extraction unit 23b extracts a second feature amount from the converted image (S353).

[0066] The processing unit acquires the output when the second feature is input to the trained object identification model 24 (S354). If the output indicates that the acquired image includes the target object (S355), the processing unit outputs information indicating that the acquired image includes the target object (S356). The information indicating that the target object is included may be an image output to a display device, or may be output as sound. Note that the processing unit may perform other processing based on the image including the target object.

[0067] In this way, by configuring the image conversion unit 22 with consideration given to the extraction of feature amounts, more appropriate conversion becomes possible, and it becomes possible for the image conversion unit 22 to absorb differences in shooting conditions such as differences in cameras.

[0068] By applying the present invention, it is possible to deal with differences in photographing conditions other than differences in cameras, such as differences in light source and ambient light during photographing.

[0069] Furthermore, the specific character strings and numerical values ​​described above and the specific character strings and numerical values ​​in the drawings are merely examples, and the present invention is not limited to these character strings and numerical values ​​and may be modified as necessary.

Claims

1. a first image acquisition unit that acquires a first image obtained by capturing an object under a first capturing condition; a second image acquisition unit that acquires a second image of the subject captured under a second photographing condition; an image conversion unit that is a machine learning model that receives the acquired second image and outputs a converted second image; an extracting unit that extracts, from a second image converted by the image converting unit, a first feature amount that indicates a feature of the first image and a second feature amount that indicates a feature of the converted second image; a transformation learning unit that causes the image transformation unit to learn based on differences in characteristics between the first feature amount and the second feature amount for a plurality of subjects; Including, the conversion learning unit causes a classifier trained with learning data including a plurality of input data each including either the first feature amount or the second feature amount and correct answer data indicating that each of the input data is either the first feature amount or the second feature amount to train the image conversion unit based on a difference between a value output when the first feature amount or the second feature amount is input and a correct answer; Information processing system.

2. 2. The information processing system according to claim 1, the first photographing condition is to photograph using a first camera, the second photographing condition is to photograph using a second camera having characteristics different from those of the first camera; Information processing system.

3. 2. The information processing system according to claim 1, a processing unit that inputs the feature amounts extracted by the extraction unit from the image captured under the second shooting condition and converted by the image conversion unit into a machine learning model that has been trained based on the feature amounts extracted by the extraction unit from the image captured under the first shooting condition, and executes processing based on an output of the machine learning model for the input feature amounts. Information processing system.

4. 4. The information processing system according to claim 3, the machine learning model outputs information indicating whether the image captured under the second shooting condition and converted by the image conversion unit includes a predetermined object. Information processing system.

5. acquiring a first image of a subject captured under a first photographing condition; acquiring a second image of the subject under a second photographing condition; extracting a first feature amount indicating a feature of the first image and a second feature amount indicating a feature of the second image converted by an image conversion unit that is a machine learning model that receives the acquired second image as input and outputs a converted second image; causing the image conversion unit to learn based on differences in characteristics between the first feature amount and the second feature amount for a plurality of objects; Including, In the step of training the image conversion unit, a classifier trained with training data including a plurality of input data each including either the first feature amount or the second feature amount and correct answer data indicating that each of the input data is either the first feature amount or the second feature amount is trained on the image conversion unit based on a difference between a value output when the first feature amount or the second feature amount is input and a correct answer. Information processing methods.

6. acquiring a first image of a subject captured under a first photographing condition; acquiring a second image of the subject under a second photographing condition; extracting a first feature amount indicating a feature of the first image and a second feature amount indicating a feature of the second image converted by an image conversion unit that is a machine learning model that receives the acquired second image as input and outputs a converted second image; causing the image conversion unit to learn based on differences in characteristics between the first feature amount and the second feature amount for a plurality of subjects; Have the computer execute the process, The process of training the image conversion unit includes a process of training the image conversion unit based on a difference between a value output when the first feature amount or the second feature amount is input to a classifier trained with training data including a plurality of input data each including either the first feature amount or the second feature amount and correct answer data indicating that each of the input data is either the first feature amount or the second feature amount, and a correct answer. program.

Images