Information processing device, information processing method, and program

Abstract

This invention provides an information processing device, an information processing method, and a program that enable users to efficiently verify the validity of the learning results of an image classification AI and implement measures to improve the accuracy of the image classification AI. [Solution] The CPU 201 of the information processing device 101 uses an image classification AI to infer a predicted label from an input image and calculate a heatmap. The CPU 201 then displays the input image, the predicted label, the predicted basis image (with the heatmap superimposed on the input image), and the correct label as a single screen on the results analysis screen of the display device 212, allowing for comparison. Subsequently, when the user presses at least one of the learning button 405, processing button 406, or delete button 407 on the results analysis screen, and then presses the export button 408, the CPU 201 updates the training data using the input image in order to retrain the image classification AI.

Description

Technical Field

[0001] The present invention relates to an information processing apparatus, an information processing method, and a program, and particularly to an information processing apparatus, an information processing method, and a program for visualizing the basis for prediction of AI.

Background Art

[0002] Techniques for visualizing the basis for prediction of AI have been proposed for the purpose of confirming the validity of learning results and improving accuracy.

[0003] For example, Non-Patent Document 1 discloses a technique for visualizing a feature region (heatmap) that is the basis for prediction of an image classification AI by superimposing it on an input image.

Prior Art Documents

Non-Patent Documents

[0004]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, with the technique of Non-Patent Document 1, although a user can determine which part of an input image the image classification AI has regarded as a feature region, the user cannot determine what learning data should be used to relearn the image classification AI for accuracy improvement based on this alone.

[0006] Therefore, an object of the present invention is to provide an information processing apparatus, an information processing method, and a program that enable a user to efficiently confirm the validity of the learning result of an image classification AI and take measures to improve the accuracy of the image classification AI. [Means for solving the problem]

[0007] To solve the above problems, the present invention provides an information processing device that inputs a force image to a trained model, causes the trained model to output a predicted label inferred from the input image, and acquires a feature region in the input image that served as the basis for the trained model's prediction, and is characterized by comprising: display means that displays the input image, the predicted label, the predicted basis image obtained by superimposing the feature region onto the input image, and a predetermined correct label indicating the correct answer to the inference for the input image on a single screen in a comparable manner; receiving means that receives an instruction to update the training data for retraining the trained model using the input image; and updating means that, upon receiving the instruction by the receiving means, updates the training data using the input image in order to retrain the trained model. [Effects of the Invention]

[0008] According to the present invention, users can efficiently verify the validity of the learning results of the image classification AI and implement measures to improve the accuracy of the image classification AI. [Brief explanation of the drawing]

[0009] [Figure 1] This figure shows an example of a system configuration of an AI prediction basis display system including an information processing device according to an embodiment of the present invention. [Figure 2] This is a block diagram showing an example of the hardware configuration of an information processing device. [Figure 3] This flowchart shows an example of a retraining / evaluation process according to an embodiment of the present invention. [Figure 4] This is the results analysis screen displayed on the display device in step S301 of Figure 3. [Figure 5] This is the training table where the paths of input images to be added to the training data are registered. [Figure 6]This is a processing table where the paths of input images to be added to or replaced in the training data after processing are registered. [Figure 7] This is a deletion table where the paths of input images to be deleted from evaluation data or training data are registered. [Figure 8] This flowchart shows an example of the data processing performed in steps S411 and S412 of Figure 4. [Figure 9] This flowchart shows another example of the data processing performed in steps S411 and S412 of Figure 4. [Figure 10] This figure shows examples of input images, prediction basis images, and processed images. [Modes for carrying out the invention]

[0010] Embodiments of the present invention will be described in detail below with reference to the drawings.

[0011] Figure 1 shows an example of the system configuration of the AI ​​prediction basis display system 1, which includes the information processing device 101 according to an embodiment of the present invention.

[0012] The AI ​​prediction basis display system 1 comprises an information processing device 101 and an external device 102, which are connected to each other via a network 110 for communication.

[0013] The information processing device 101 is a device operated by the user, and displays the input images and associated data that are subject to the retraining / evaluation process (Figure 3) described later on the results analysis screen (Figure 4), and accepts the user pressing various buttons on this screen.

[0014] The external device 102 manages the input images and associated data that are subject to processing in the above-mentioned retraining / evaluation process.

[0015] In the embodiments of the present invention, the information processing apparatus 101 will be described as executing the processes shown in the flowchart of FIG. 3, but an embodiment in which the external apparatus 102 executes the processes may also be used. Further, regarding the processes described as being performed by the external apparatus 102, such as the management of the images to be processed, the information processing apparatus 101 may perform these processes.

[0016] FIG. 2 is a block diagram showing an example of the hardware configuration of the information processing apparatus 101. Note that since the external apparatus 102 has the same hardware configuration as the information processing apparatus 101, duplicate descriptions will be omitted.

[0017] In FIG. 2, the information processing apparatus 101 includes a CPU 201, a ROM 202, a RAM 203, a storage device 204, an input control unit 205, an audio control unit 206, a video control unit 207, a memory control unit 208, and a communication I / F control unit 209. These devices and controllers are connected to each other via a system bus 200.

[0018] The CPU 201 comprehensively controls each device and controller connected to the system bus 200.

[0019] The RAM 203 functions as the main memory, work area, etc. of the CPU 201. When executing a process, the CPU 201 loads a program or the like necessary for the execution of the process from the ROM 202 or an external memory 213 into the RAM 203, and realizes various operations by executing the loaded program.

[0020] The storage device 204 includes an SSD, an HDD, etc., and holds an image classification AI or the like, which will be described later.

[0021] The input control unit 205 controls input from the input device 210, which consists of a keyboard, touch panel, mouse, or other pointing device. For example, if the input device 210 is a touch panel, the user can give various instructions by pressing (touching with a finger, etc.) icons, cursors, or buttons displayed on the touch panel. In this case, the touch panel may be a multi-touch screen or other touch panel capable of detecting the positions of multiple fingers touching it.

[0022] The audio control unit 206 controls the input and output of audio to audio input / output devices 211 such as microphones and speakers.

[0023] The video control unit 207 controls the display on the display device 212, which consists of a display or projector. In this case, the display includes the display of a notebook computer integrated with the main unit. Furthermore, if the display device 212 is a device that can accept touch operations as described above, it also serves as an input device 210. The video control unit 207 can also control the video memory (VRAM) for display control, and may use a portion of the RAM 203 as the video memory area, or a separate dedicated video memory may be provided.

[0024] The memory control unit 208 controls access to the external memory 213. The external memory 213 is not particularly limited as long as it is a storage device that stores boot programs, various applications, font data, user files, editing files, and various other data. For example, external storage devices (hard disks), flexible disks (floppy disks), or CompactFlash® memory connected to a PCMCIA card slot via an adapter can be used as the external memory 213.

[0025] The communication interface control unit 209 connects to and communicates with external devices via the network 110 and executes communication control processing on the network 110. For example, it can communicate using TCP / IP, telephone lines such as ISDN, and 4G and 5G mobile phone lines.

[0026] Furthermore, the CPU 201 enables display on the display device 212 by, for example, performing the process of expanding (rasterizing) outline fonts into the display information area in RAM 203. The CPU 201 can also accept user commands via a mouse cursor (not shown) on the display device 212.

[0027] Next, the retraining / evaluation process according to an embodiment of the present invention will be explained using the flowchart in Figure 3.

[0028] This process is executed when the CPU 201 of the information processing device 101 reads a predetermined control program. In this process, information regarding the prediction results of the image classification AI is displayed on the results analysis screen (Figure 4). Furthermore, after the training data and evaluation data are updated in response to operations received from the user on the results analysis screen (pressing the training, processing, or delete buttons), the image classification AI is retrained and evaluated using the updated training data and evaluation data. The training data and evaluation data will be described later.

[0029] First, in step S301, the CPU 201 acquires the input image and associated data from the external device 102, specifically the correct label, predicted label, and heatmap information, and stores them in the RAM 203. Subsequently, the CPU 201 controls the display device 212 to display the correct label, predicted label, input image, and prediction basis image on the result analysis screen (Figure 4: display means) based on the input image and associated data.

[0030] Here, the correct label is a label predetermined by the user or others that indicates the correct answer in the image classification AI's inference for the input image.

[0031] The image classification AI is a pre-trained model using a Convolutional Neural Network (CNN) that is pre-trained in the learning unit (not shown in Figure 1) within the external device 102 and stored in the memory device 204. While a CNN is used here for the image classification AI, it is not limited to this; other known algorithms such as support vector machines may also be used.

[0032] When the CPU 201 receives an input image (and its correct label) from an external source, it reads the image classification AI from the storage device 204 and inputs the received input image into the read image classification AI, thereby outputting a predicted label inferred from the input image. The CPU 201 transmits the input image, along with the associated data, the correct label and the predicted label, to the external device 102 in advance of this process.

[0033] The heatmap represents the feature regions in the input image that formed the basis for the image classification AI's predictions. The CPU 201 calculates the heatmap using Grad-CAM, a well-known technique. The CPU 201 transmits this heatmap information to the external device 102 before this process, as information linked to the input image. While Grad-CAM is used here to calculate the heatmap, the process is not limited to this; other known algorithms such as RISE, Score-CAM, XRAI, and Group-CAM may also be used.

[0034] Thus, the CPU 201 stores the input image, its associated correct label, predicted label, and heatmap information in the external device 102 before the main process. The input image is managed in a way that allows for identification of whether it is training data or evaluation data, such as by registering it with information indicating whether it is training data or evaluation data. Furthermore, although the CPU 201 calculates the heatmap in this embodiment, the CPU 201 is not limited to this, as long as the heatmap information is stored in the external device 102 before the main process. For example, the CPU 201 may acquire a heatmap calculated externally.

[0035] The prediction basis image is an image that visualizes the prediction basis of the image classification AI by overlaying a heatmap onto the input image. In the prediction basis image, the heatmap is color-coded. Specifically, the higher the degree of prediction basis, the warmer the color of the heatmap's feature area, and the lower the degree of prediction basis, the cooler the color of the heatmap's feature area. For example, the heatmap's feature areas are colored red, orange, yellow, yellow-green, and light blue in order of decreasing degree of prediction basis.

[0036] In step S302, the CPU 201 determines whether the input image (target) received in step S301 is evaluation data. If the result of this determination is that the target is evaluation data (YES in step S302), the process proceeds to step S303; otherwise, the process proceeds to step S305.

[0037] In step S303, the CPU 201 determines whether the learning button 405 on the results analysis screen in Figure 4 was pressed. If the learning button 405 was pressed (YES in step S303), the process proceeds to step S304; otherwise, it proceeds to step S305.

[0038] In step S304, the CPU 201 registers the path of the target input image in the learning table (Figure 5) in order to improve the accuracy of the image classification AI by retraining it using the target input image and its correct label. For example, if the predicted label of the target is incorrect, retraining the image classification AI using the target input image and its correct label is expected to prevent similar errors from occurring again. Compared to the data processing described later, this method can correct the predicted label output by the image classification AI regardless of the original prediction basis, making it suitable for cases where the prediction basis is correct but the predicted label is incorrect.

[0039] In step S305, the CPU 201 determines whether the processing button 406 on the results analysis screen in Figure 4 was pressed. If the processing button 406 was pressed (YES in step S305), the process proceeds to step S306; otherwise, it proceeds to step S307.

[0040] In step S306, the CPU 201 registers the path of the target input image in the processing table (Figure 6) in order to perform data processing. Data processing refers to the process of replacing inappropriate parts in the input image with a single value if those inappropriate parts become feature parts that serve as the basis for prediction.

[0041] Here, the data processing exemplified in Figure 8, described later, is intended to be performed on all input images whose paths are registered in the processing table when the export button 408 (Figure 4) is pressed. The data processing exemplified in Figure 9, described later, allows the user to set the probability of processing in order to prevent the consistency of the processing itself from being used as the basis for prediction, and is intended to be performed on the input images used for retraining the image classification AI. In this way, if the basis for prediction is inappropriate, it is expected that by processing the input image and then using it to retrain the image classification AI, the image classification AI will learn the correct basis for prediction and make correct predictions. Compared to the retraining process described above, the data processing can correct the basis for prediction, so it is considered suitable when the predicted label is also wrong because the basis for prediction is wrong.

[0042] In step S307, the CPU 201 determines whether the delete button 407 on the results analysis screen in Figure 4 was pressed. If the delete button 407 was pressed (YES in step S307), the process proceeds to step S308; otherwise, it proceeds to step S309. Here, the delete button 407 is pressed when the user wants to delete an input image from the training data or evaluation data because using that input image as training data or evaluation data would negatively affect the retraining or evaluation of the image classification AI. Here, an example of an input image negatively affecting the retraining or evaluation of the image classification AI is when the input image included in the input information is an image unrelated to training or evaluation, or when the object is too large or too small.

[0043] In step S308, CPU201 registers the path of the target input image in the deletion table (Figure 7) in order to execute the above deletion process.

[0044] In step S309, the CPU 201 determines whether the export button 408 on the results analysis screen in Figure 4 was pressed. If the export button 408 was pressed (YES in step S309), the process proceeds to step S310; otherwise, it returns to step S301.

[0045] In step S310, the CPU 201 determines whether the target input image is evaluation data or not. If the result of this determination is that the target is evaluation data (YES in step S310), the process proceeds to step S311; otherwise, i.e., the target is training data (NO in step S310), the process proceeds to step S312.

[0046] In step S311, if the path to the target input image is registered in the learning table, the CPU 201 (update means) changes the target input image from evaluation data to learning data. That is, if the learning result display unit 400 (third receiving means) receives confirmation that the export button 408 has been pressed after the learning button 405 has been pressed, the storage location of the target input image is moved from the evaluation data folder of the external device 102 to the learning data folder. Also, if the path to the target input image is registered in the processing table, the CPU 201 processes the target input image using data processing and changes the processed input image from evaluation data to learning data. That is, if the learning result display unit 400 (first receiving means) receives confirmation that the export button 408 has been pressed after the processing button 406 has been pressed, the storage location of the target input image after processing is moved from the evaluation data folder of the external device 102 to the learning data folder. Furthermore, if the path to the target input image is registered in the deletion table, the CPU 201 deletes the target input image from the evaluation data. In other words, if the learning result display unit 400 (second reception means) receives a press of the export button 408 after the delete button 407 has been pressed, the target input image is deleted from the evaluation data folder of the external device 102 where it is currently stored. Then, the process proceeds to step S313.

[0047] Thus, when the target is evaluation data, not only the evaluation data but also the training data is updated. However, in this case, if only the delete button 407 is pressed and then the export button 408 is pressed, the training data will not be updated, and therefore, retraining of the image classification AI in step S313 described later is not necessary.

[0048] In step S312, if the path to the target input image is registered in the processing table, the CPU 201 (update means) processes the target input image through data processing and replaces the target input image as training data with the processed input image. That is, it replaces the target input image in the training data folder of the external device 102 with the processed target input image. Also, if the path to the target input image is registered in the deletion table, the CPU 201 deletes the target input image from the training data. That is, it deletes the target input image from the training data folder of the external device 102 where it is currently stored. After that, the process proceeds to step S313.

[0049] Thus, when the target is training data, only the training data is updated, and the evaluation data is not updated. For this reason, when the target is training data, it is not necessary to evaluate the image classification AI in step S313 described later.

[0050] In step S313, CPU 201 outputs training data from the training data folder of external device 102 and retrains the image classification AI. CPU 201 also outputs evaluation data from the evaluation data folder of external device 102 and evaluates the image classification AI. Afterward, this process terminates.

[0051] In this embodiment, the retraining of the image classification AI was performed by the CPU 201 of the information processing device 101. However, the embodiment is not limited to this one, as long as the retraining of the image classification AI is performed using updated training data. For example, the CPU 201 may instruct the external device 102 to perform retraining of the image classification AI and obtain the retrained image classification AI (or its internal parameters) from the external device 102. Similarly, the evaluation of the image classification AI was performed by the CPU 201 of the information processing device 101. However, the embodiment is not limited to this one, as long as the evaluation of the image classification AI is performed using updated evaluation data. For example, the CPU 201 may instruct the external device 102 to perform evaluation of the image classification AI and obtain the evaluation results from the external device 102.

[0052] Figure 4 shows the results analysis screen displayed on the display device 212 in step S301 of Figure 3.

[0053] The results analysis screen includes learning result display sections 400a to 400d for each input image (hereinafter collectively referred to as "learning result display section 400"), and an export button 408.

[0054] The learning result display unit 400 consists of a correct label display unit 401, a predicted label display unit 402, an input image display unit 403, a prediction basis image display unit 404, a learning button 405, a processing button 406, and a delete button 407.

[0055] The correct label display unit 401 displays the correct label for the input image.

[0056] The prediction label display unit 402 displays the prediction label of the image classification AI.

[0057] The input image display unit 403 displays the input image.

[0058] The prediction basis image display unit 404 displays the prediction basis image.

[0059] As shown in Figure 4, the correct label, predicted label, input image, and prediction basis image are all displayed side-by-side on a single screen, the results analysis screen. This allows users to easily judge the validity of the learning results using the results analysis screen in Figure 4.

[0060] The learning button 405 (reception mechanism) is a button that accepts instructions to register the path of the input image in the learning table.

[0061] The processing button 406 (reception mechanism) is a button that accepts instructions to register the path of the input image in the processing table.

[0062] The delete button 407 (reception method) is a button that accepts instructions to register the path of the input image in the delete table.

[0063] The export button 408 (reception mechanism) is a button that accepts instructions to output training data and evaluation data after processing the input images registered in the training table, processing table, and deletion table.

[0064] If the user determines that the learning results displayed on the learning results display unit 400 are not valid, they press at least one of the learning button 405, processing button 406, or delete button 407 on the results analysis screen in Figure 4, and then press the export button 408. By simply operating the learning results display unit 400 (first receiving means, second receiving means, third receiving means), the user can easily implement measures to improve the accuracy of the image classification AI. Specific accuracy improvement measures include adding input images as learning data, processing input images for use as learning data and evaluation data, and deleting input images from the learning data and evaluation data.

[0065] Figure 5 shows the training table where the paths of input images to be added to the training data are registered.

[0066] Figure 6 is a processing table in which the paths of input images to be added to or replaced in the training data after processing are registered.

[0067] Figure 7 is a deletion table in which the paths of input images to be deleted from evaluation data or training data are registered.

[0068] Next, using the flowchart in Figure 8, an example of the data processing performed in steps S411 and S412 of Figure 4 will be explained. This process is performed on all input images for which a path has been registered in the processing table when the export button 408 (Figure 4) is pressed. In other words, in this process, all input images for which a path has been registered in the processing table are processed so that the image classification AI is not trained based on incorrect prediction criteria.

[0069] This process is executed when the CPU 201 of the information processing device 101 reads a predetermined control program.

[0070] First, in step S801, the CPU201 selects one of the input images whose path is registered in the processing table as the processing target. An example of an input image selected as the processing target is shown in Figure 10(A). In this input image, the position of the left hole is deviated from the design position, so its correct label is "abnormal". Therefore, it is desirable that the predicted label inferred by the image classification AI from this input image is also "abnormal". Furthermore, it is desirable that the heatmap of the prediction basis image be colored with warmer colors (colors indicating a higher degree of prediction basis) the closer it is to the position of the left hole.

[0071] In step S802, the CPU 201 obtains the coordinates of pixels in the heatmap that have a value greater than the set value 1A. Each pixel in the heatmap takes a value between 0 and 1. The set value 1A is a threshold value used to determine whether the degree to which each pixel in the heatmap was used as the basis for prediction meets a predetermined criterion, and it has a value greater than 0 and less than 1. An example of a prediction basis image is shown in Figure 10(B). In this prediction basis image, the area around the hole on the right is used as the prediction basis, and it can be seen that the prediction label was output based on an incorrect prediction basis. Step S802 processes such images.

[0072] In step S803, the CPU 201 performs a process in which it replaces the pixel values ​​of all coordinates obtained in step S802 in the input image with the set value 2A. Figure 10(C) shows the image (processed image) after the processing in step S803 has been performed on the input image in Figure 10(A). In this processed image, black is specified as the set value 2A and the processing is carried out.

[0073] In step S804, the CPU 201 determines whether the processing in step S803 has been completed for all input images for which a path is registered in the processing table. If the processing in step S803 has been completed for all input images (YES in step S804), the process is terminated; otherwise (NO in step S804), the process returns to step S801.

[0074] Next, using the flowchart in Figure 9, another example of the data processing performed in steps S411 and S412 of Figure 4 will be explained. In this process, the processing of the input images is performed with a certain probability in order to prevent the consistency of the processing of the input images themselves from affecting the calculation of the heatmap. Furthermore, when the export button 408 (Figure 4) is pressed, this processing is performed probabilistically not only on the input images whose paths are registered in the processing table, but also on all input images stored in the external device 102. This process is particularly intended to be applied during additional learning. That is, in this process, the processing of all input images in the information processing device 101 is applied probabilistically in order to prevent the image classification AI from being trained based on incorrect prediction basis.

[0075] This process is executed when the CPU 201 of the information processing device 101 reads a predetermined control program.

[0076] First, in step S901, the CPU 201 acquires one of the input images stored in the external device 102.

[0077] In step S902, the CPU 201 determines whether the path of the input image acquired in step S901 is registered in the processing table. If it is registered in the processing table (YES in step S902), the process proceeds to step S903; otherwise (NO in step S902), the process proceeds to step S906.

[0078] In step S903, the CPU 201 generates a random number and determines whether the generated random number is greater than the set value 1B. Both the generated random number and the set value 1B take values ​​between 0 and 1. If the result of the determination is that the random number is greater than the set value 1B (YES in step S903), the process proceeds to step S904 to process the input image acquired in step S901; otherwise (NO in step S903), the process ends. By applying the processing probabilistically in this way, it is expected that the consistency of the processing itself will not affect the calculation of the heatmap.

[0079] In step S904, the CPU 201 obtains the coordinates of pixels in the heatmap that have a value greater than the set value 2B, and proceeds to step S905. Each pixel in the heatmap takes a value between 0 and 1. The set value 2B is a threshold value used to determine whether the degree that formed the basis of the prediction for each pixel in the heatmap meets a predetermined standard, and has a value greater than 0 and less than 1.

[0080] In step S905, the CPU 201 performs a process in which it replaces the pixel values ​​of all coordinates obtained in step S904 with the set value 3B in the input image obtained in step S901, and then terminates this process.

[0081] In step S906, the CPU 201 generates a random number and determines whether the generated random number is greater than the set value 1B. If the result of the determination is that the random number is greater than the set value 1B (YES in step S906), the process proceeds to step S907 to process the input image acquired in step S901; otherwise (NO in step S906), the process ends. In this way, by probabilistically processing input images stored in the external device 102 that do not have a path registered in the processing table, it is expected that the consistency of the processing itself will not affect the calculation of the heatmap.

[0082] In step S907, the CPU 201 randomly selects pixels from the input image acquired in step S901, replaces the value of the randomly selected pixels with the set value 3B, and then terminates the process. The number of pixels randomly selected here may be a randomly set value or a fixed value.

[0083] (Other embodiments) In this embodiment, the system can also be implemented by supplying a program that implements one or more functions to a computer of a system or device via a network or storage medium, and the system control unit of that system or device reads and executes the program. The system control unit has one or more processors or circuits and may include a plurality of separate system control units or a network of a plurality of separate processors or circuits in order to read and execute executable instructions.

[0084] A processor or circuit may include a central processing unit (CPU), a microprocessing unit (MPU), a graphics processing unit (GPU), an application-specific integrated circuit (ASIC), or a field-programmable gate array (FPGA). Alternatively, a processor or circuit may include a digital signal processor (DSP), a dataflow processor (DFP), or a neural processing unit (NPU).

[0085] Although preferred embodiments of the present invention have been described above, the present invention is not limited to these embodiments, and various modifications and changes are possible within the scope of its gist. [Explanation of symbols]

[0086] 1. AI Prediction Basis Display System 101 Information Processing Device 102 External device 110 Network 201 CPU 212 Display device 405 Learning Button 406 Processing Button 407 Delete button 408 Export button

Claims

1. An information processing device that inputs an input image into a trained model, causes the trained model to output a predicted label inferred from the input image, and acquires the feature region in the input image that served as the basis for the trained model's prediction, A display means that displays the input image, the prediction label, the prediction basis image obtained by superimposing the feature region onto the input image, and a predetermined correct label indicating the correct inference for the input image, in a comparable manner on a single screen. A receiving means for receiving instructions to update training data for retraining the trained model using the input image, Upon receiving the instruction via the reception means, an update means updates the training data using the input image in order to retrain the trained model, An information processing device characterized by comprising:

2. The information processing apparatus according to claim 1, characterized in that the receiving means includes a first receiving means that receives an instruction to use the input image as training data after processing it.

3. The information processing apparatus according to claim 2, characterized in that, after acquiring the input image as learning data, the update means, upon receiving an instruction from the first receiving means, replaces the input image in the learning data with the processed input image and updates the learning data.

4. The information processing apparatus according to claim 2, characterized in that, after the update means acquires the input image as evaluation data for evaluating the trained model, and then receives an instruction from the first receiving means, it moves the processed input image from the evaluation data to the training data and updates the training data and the evaluation data in order to retrain and evaluate the trained model.

5. The information processing apparatus according to any one of claims 1 to 4, characterized in that the receiving means includes a second receiving means for receiving an instruction that the input image will not be used as training data.

6. The information processing apparatus according to claim 5, characterized in that the update means, after acquiring the input image as learning data, receives an instruction from the second receiving means, deletes the input image from the learning data, and updates the learning data.

7. The information processing apparatus according to claim 5, characterized in that, after the update means acquires the input image as evaluation data for evaluating the trained model, and then receives an instruction from the second receiving means, it deletes the input image from the evaluation data and updates the evaluation data in order to evaluate the trained model.

8. The receiving means includes a third receiving means that receives an instruction to use the input image as training data when the input image has been acquired as evaluation data for evaluating the trained model, The information processing apparatus according to any one of claims 1 to 7, characterized in that when the update means receives an instruction from the third receiving means, it moves the input image from the evaluation data to the training data and updates the training data and the evaluation data in order to retrain and evaluate the trained model.

9. An information processing method for an information processing device, which inputs an input image into a trained model, causes the trained model to output a predicted label inferred from the input image, and obtains the feature region in the input image that served as the basis for the trained model's prediction, wherein A display step that displays the input image, the prediction label, the prediction basis image obtained by superimposing the feature region onto the input image, and a predetermined ground truth label indicating the correct inference for the input image on a single screen in a comparable manner. A receiving step that receives an instruction to update the training data for retraining the trained model using the input image, In response to the instructions received in the reception step, an update step is performed to update the training data using the input image in order to retrain the trained model, An information processing method characterized by having the following features.

10. A computer-executable program that causes a computer to function as one of the means of an information processing device according to any one of claims 1 to 8.

Images