JP7367383B2 - Image recognition application construction support system and program - Google Patents

Description

The present invention relates to a system (hereinafter referred to as "image recognition application construction support system") and a program for supporting construction of an image recognition application.

In recent years, as the performance of image recognition has improved through deep learning, the field application of applications that utilize image recognition is being considered in various fields. In environments where the positional relationship between the subject and the camera is restricted to some extent, such as facial recognition at airport gates and inspection of industrial products on factory production lines, on-site applications of applications that utilize image recognition have begun. ing.

If the viewpoint of the camera that photographs the subject is fixed, it is easy to use existing open datasets such as ImageNet for learning, and it is also easy to add data at the installation site. An example of a case where the viewpoint of a camera that photographs a subject is fixed is face recognition, where the viewpoint is in front of the user.

If the viewpoint of the camera that photographs the subject cannot be fixed, it is necessary to prepare training data equal to the number of assumed viewpoints. However, existing open datasets have viewpoint bias, and models trained on datasets with viewpoint bias will have variations in recognition accuracy depending on the viewpoint.

Note that when preparing a shape model to compensate for the lack of learning data, it is necessary to create a shape model. However, since creating a shape model requires a high level of skill, it is difficult for general users to create such a shape model.

On the other hand, images taken from various viewpoints may be prepared without creating a shape model. However, in this case, creating a dataset requires a huge amount of time.

JP 2018-55195 Publication JP2017-102838A Japanese Patent Application Publication No. 2014-157509 Japanese Unexamined Patent Publication No. 1-136283

With conventional image recognition application construction support systems, it is difficult to collect data without waste and reduce the number of man-hours required for construction work.

Therefore, one aspect of the present invention is to provide an image recognition application construction support system and program that can collect data without waste and reduce the number of man-hours for construction work.

According to one proposal, it is an image recognition application construction support system that can wirelessly communicate with a terminal device, which stores and manages meta information indicating from which viewpoint each image of an image data set was taken in memory together with the image. a management means for statistically analyzing the distribution of viewpoints of a plurality of images used by the image recognition engine for model learning among the information stored in the memory; Construction of an image recognition application, comprising: presentation means for presenting an image of a missing viewpoint in an identifiable manner to the terminal device by presenting it as a heat map of the model representing the distribution of values counted for each viewpoint coordinate ; Support systems are provided.

According to one aspect, data can be collected without waste and the number of man-hours for construction work can be reduced.

FIG. 1 is a functional block diagram showing an example of an image recognition application construction support system in a first embodiment. 2 is a flowchart illustrating an example of the operation of the image recognition application construction support system in the first embodiment. FIG. 1 is a block diagram illustrating an example of a computer system. FIG. 2 is a functional block diagram showing an example of an image recognition application construction support system in a second embodiment. It is a flow chart explaining an example of operation of an image recognition application construction support system in a 2nd example. FIG. 2 is a diagram illustrating the effects of the image recognition application construction support system.

Embodiments of the disclosed image recognition application construction support system and program will be described below with reference to the drawings.

FIG. 1 is a functional block diagram showing an example of an image recognition application construction support system in the first embodiment. The image recognition application construction support system (hereinafter also simply referred to as "support system") 1 shown in FIG. It has a section 14.

The viewpoint coordinate extraction unit 3 acquires an image from the image data set 2 and extracts the viewpoint coordinates of the camera that photographed the object in the image (hereinafter also referred to as "image viewpoint coordinates") using a well-known method. Extract with The well-known method of extracting the viewpoint coordinates of the camera that photographed the object in the image is not particularly limited, but for example, the well-known technique described in Patent Document 4 may be used. The image data set 2 may be an existing open data set such as ImageNet. The viewpoint coordinate extraction unit 3 supplies the acquired image and the viewpoint coordinates of the extracted image to the management unit 11 of the support system 1. The image data set 2 and the viewpoint coordinate extraction unit 3 may be included in a cloud computing system that can be accessed from the support system 1. Further, the viewpoint coordinate extraction unit 3 may be included in the support system 1.

The terminal device 4 can be formed by a portable electronic device such as a smartphone or a tablet. The terminal device 4 includes a communication device that communicates with the support system 1 and an input device. The communication device of the terminal device 4 has a wireless communication function that communicates with the support system 1 using, for example, signals of the Wi-Fi (registered trademark) standard. Note that the wireless communication function may perform communication using signals of standards other than Wi-Fi (registered trademark), such as BT (Bluetooth (registered trademark)) and BLE (Bluetooth (registered trademark) Low Energy). That goes without saying. The input device of the terminal device 4 includes a camera, a sensor that outputs information (hereinafter also referred to as "sensor information") used to estimate the tilt (or posture) of the camera using a well-known method, and the like. Known methods for estimating camera tilt from sensor information are not particularly limited.

In the support system 1, the management unit 11 stores one or more images used for learning the image recognition engine obtained from the viewpoint coordinate extraction unit 3 and the viewpoint coordinates of the images in a database (DB) as meta information. ) is an example of a management means for storing and managing the data. The management unit 11 associates each image of the learning image data set with the viewpoint coordinates of each image, and stores and manages the linked information in a DB as meta information. That is, the management unit 11 stores and manages meta information indicating from which viewpoint each image of the image data set 2 was photographed in the DB together with the images. The DB can be formed using a memory within the support system 1, for example.

The analysis unit 12 is an example of an analysis unit that statistically analyzes the distribution of viewpoints of a plurality of images used by the image recognition engine for model learning, among the information stored in the DB. The analysis unit 12 statistically analyzes the distribution of viewpoints of images used for learning for each learned model. The presentation unit 13 is an example of presentation means that presents the results of the statistical analysis to the terminal device 4 in the form of numerical values, diagrams, etc. The presentation unit 13 displays, for example, the results of the statistical analysis on the display device of the terminal device 14.

When the estimation unit 14 takes an image with the camera of the terminal device 4 and adds it as learning image data, the estimation unit 14 determines whether the camera of the terminal device 4 photographs the object based on the sensor information obtained from the sensor of the terminal device 4. This is an example of an estimating means for estimating the viewpoint coordinates using a well-known method. The viewpoint coordinates of an image can be estimated by a known method from the positional relationship between the tilt of the camera estimated by a known method from sensor information and, for example, an object in the center of the image.

FIG. 2 is a flowchart illustrating an example of the operation of the image recognition application construction support system in the first embodiment. Among the processes shown in FIG. 2, the process in step S1 can be executed by the management unit 11. The process in step S2 can be executed by the analysis unit 12, and the process in step S3 can be executed by the presentation unit 13. The processing in steps S4 and S5 can be executed by the terminal device 4, and the processing in steps S6 and S7 can be executed by the estimation unit 14.

In step S1, the management unit 11 links one or more images of the training image dataset used for learning the image recognition engine, obtained from the viewpoint coordinate extraction unit 3, with the viewpoint coordinates of the images, and metadata Save and manage information in a DB. A DB that associates one or more images with viewpoint coordinates and stores them as meta information can be formed in the memory within the support system 1, for example.

In step S2, the analysis unit 12 statistically analyzes the distribution of viewpoints of a plurality of images used by the image recognition engine for model learning among the information stored in the DB.

In step S3, the presentation unit 13 presents the results of the statistical analysis to the terminal device 4 in the form of numerical values, diagrams, etc. When presenting the results of the statistical analysis in a diagram, for example, a heat map of the model may be displayed on the display device of the terminal device 4. In this case, the heat map of the model represents, for example, the distribution of values obtained by counting the viewpoints of each image for each viewpoint coordinate. Therefore, by presenting the results of the statistical analysis as a heat map of the model, the presentation unit 13 determines whether or not the value counted for each viewpoint coordinate of each image is less than a certain value. Viewpoint images can be presented in a distinguishable manner.

In step S4, the terminal device 4 determines whether the distribution of viewpoints is sufficient, based on, for example, the presented statistical analysis results. Whether the distribution of viewpoints is sufficient may be determined, for example, based on whether or not a value obtained by counting the viewpoints of each image for each viewpoint coordinate is equal to or greater than a certain value in a heat map. If the terminal device 4 determines that the distribution of viewpoints is not sufficient and the determination result in step S4 is NO, the process proceeds to step S5, which will be described later. On the other hand, if the terminal device 4 determines that the viewpoint distribution is sufficient and the determination result in step S4 is YES, the process proceeds to step S8, the image recognition engine relearns the model, and the process ends. .

In step S5, the terminal device 4 notifies the user by displaying a message urging the user to take an image from the missing viewpoint coordinates based on the result of the statistical analysis. When the user photographs images from the missing viewpoint coordinates with the camera of the terminal device 4 in response to the notified message, in step S5, the terminal device 4 acquires sensor information at the time of photographing each image. It is sent to the support system 1 along with the image. Specifically, the acquired sensor information is transmitted to the management unit 11 of the support system 1 together with the image.

When the terminal device 4 executes the determination in step S4 and the terminal device 4 side takes the initiative in photographing in step S5 and has the user take the lead, the user basically responds to the instructions from the terminal device 4. It is convenient because all you have to do is perform the operation.

Note that in step S4, the user may look at a heat map or the like displayed on the display device of the terminal device 4 and judge whether the distribution of viewpoints is sufficient. In this case, the user can easily understand the distribution of viewpoints from the heat map displayed on the display device of the terminal device 4. If the user determines that the distribution of viewpoints is insufficient and selects a transmission operation to transmit the missing information, the determination result in step S4 becomes NO, and the process proceeds to step S5. In step S5, when the user shoots an image from the missing viewpoint coordinates with the camera of the terminal device 4 and obtains the sensor information at the time of shooting each image from the sensor of the terminal device 4, the user supports the image and sensor information. It is transmitted to the estimation unit 14 of the system 1. On the other hand, if the user determines that the viewpoint distribution is sufficient and does not select the above transmission operation, the determination result in step S4 becomes YES, and the process proceeds to step S8, where the image recognition engine relearns the model. , the process ends. For example, if the user does not select a sending operation within a certain period of time, it may be determined that the viewpoint distribution is sufficient, and the determination result in step S4 may be YES.

If the user takes the initiative in making the determination in step S4 and taking the photograph in step S5, the processing load on the terminal device 4 can be reduced.

In step S6, the estimation unit 14 of the support system 1 estimates the viewpoint coordinates of the image received from the terminal device 4 using a well-known method based on the sensor information received from the terminal device 4. Specifically, the viewpoint coordinates of the image received from the terminal device 4 are determined using a well-known method based on the tilt of the camera estimated from sensor information using a well-known method and the positional relationship with, for example, an object in the center of the image. Estimate by.

In step S7, the estimation unit 14 instructs the management unit 11 to link the photographed image and the estimated viewpoint coordinates and add them to the DB of the support system 1 as meta information, and the process returns to step S2. As a result, in response to the instruction from the estimation unit 14, the management unit 11 links the image taken from the missing viewpoint coordinate with the viewpoint coordinate when the object in the image was photographed, and provides support. Add it to the system 1 DB.

FIG. 3 is a block diagram showing an example of a computer system. The computer system 100 having the hardware configuration shown in FIG. 3 can form, for example, the support system 1 shown in FIG. 1.

Computer system 100 includes a processor 121 , a memory 122 , an input device 123 , a display device 124 , and an interface (or communication device) 125 that are connected to each other via a bus 120 . The processor 121 can be formed of, for example, a central processing unit (CPU), and executes programs stored in the memory 122 to control the entire computer system 100.

The memory 122 can be formed of, for example, a semiconductor storage device, a magnetic recording medium, an optical recording medium, a magneto-optical recording medium, or the like. The memory 122 is an example of a storage device that stores various programs executed by the processor 121, various data, etc., and can form a DB of the support system 1. The storage device may be a portable recording medium such as a CD-ROM (Compact Disk-Read Only Memory), a DVD (Digital Versatile Disk), or a USB (Universal Serial Bus) memory, or a semiconductor memory such as a flash memory. Furthermore, the storage device may be a computer-readable storage medium that stores a program. Furthermore, the storage device may be formed of a plurality of storage devices, and in this case, may include a storage device externally connected to the computer system 100.

The input device 123 can be formed of, for example, a keyboard, which is operated by a user, and is used to input commands and data to the processor 121. The display device 124 is used to display messages, processing results, etc. to the user. Note that a touch panel in which the input device 123 and the display device 124 are integrated may be used. The interface 125 is an example of a communication means that has a wireless communication function and manages the exchange of information, commands, etc. between the computer system 100 and other devices.

When the computer system 100 forms the support system 1 shown in FIG. 1, the processor 121 executes the program stored in the memory 122 to execute steps S1, S2, S3, S6, and S7 shown in FIG. .

Note that the computer system 100 is not limited to a hardware configuration in which the components of the computer system 100 are connected via the bus 120, and can be formed by a general-purpose computer.

Further, the computer system 100 having the hardware configuration shown in FIG. 3 can form, for example, the terminal device 4 shown in FIG. 1. In the case of the computer system 100 forming the terminal device 4, the input device 123 further includes a camera, a sensor outputting information used to estimate the tilt of the camera, and the like. The sensor that outputs information used to estimate the tilt of the camera may be, for example, a three-axis acceleration sensor, a gyro sensor, or the like.

When the computer system 100 forms the terminal device 4 shown in FIG. 1, the processor 121 executes the program stored in the memory 122 to execute the processes of steps S4 and S5 shown in FIG.

FIG. 4 is a functional block diagram showing an example of an image recognition application construction support system in the second embodiment. In FIG. 4, the same parts as in FIG. 1 are designated by the same reference numerals, and their explanation will be omitted. The support system 1A shown in FIG. 4 includes a management section 11, an analysis section 12, and a presentation section 13.

The terminal device 4A can be formed by, for example, a smartphone, a tablet, or the like. The terminal device 4 includes a communication device that communicates with the support system 1A, an input device including a camera and a sensor, and an estimator 14A. When the estimation unit 14A takes an image with the camera of the terminal device 4A and adds it as learning image data, the estimation unit 14A determines whether the camera of the terminal device 4A photographs the object based on the sensor information obtained from the sensor of the terminal device 4A. This is an example of an estimating means for estimating the viewpoint coordinates using a well-known method. The viewpoint coordinates of an image can be estimated by a known method from the positional relationship between the tilt of the camera estimated by a known method from sensor information and, for example, an object in the center of the image.

FIG. 5 is a flowchart illustrating an example of the operation of the image recognition application construction support system in the second embodiment. In FIG. 5, steps that are the same as those in FIG. 2 are given the same reference numerals, and their explanations will be omitted.

Among the processes shown in FIG. 5, the processes in steps S4 and S5 can be executed by the terminal device 4A, and the processes in steps S6A and S7A can be executed by the estimation unit 14A of the terminal device 4A.

In steps S4 and S5, the terminal device 4A performs the same processing as the terminal device 4 in the first embodiment. In step S6A, the estimating unit 14A of the terminal device 4A estimates the viewpoint coordinates when the object in the image was photographed using a well-known method based on the sensor information received from the sensor of the terminal device 4A. In step S7A, the estimation unit 14A of the terminal device 4A instructs the management unit 11 to link the photographed image and the estimated viewpoint coordinates and add them as meta information to the DB of the support system 1A, and the process continues in step S7A. Return to S2. As a result, in response to the instruction from the estimation unit 14A, the management unit 11 links the image taken from the missing viewpoint coordinate with the viewpoint coordinate when the object in the image was photographed, and provides support. Add it to the DB of system 1A.

When the computer system 100 shown in FIG. 3 forms the support system 1A shown in FIG. 4, the processor 121 executes the program stored in the memory 122 to execute the processes of steps S1, S2, and S3 shown in FIG. .

When the computer system 100 shown in FIG. 3 forms the terminal device 4A shown in FIG. Execute.

FIG. 6 is a diagram illustrating the effects of the image recognition application construction support system. In FIG. 6, the same parts as in FIGS. 1 and 4 are given the same reference numerals, and their explanations will be omitted.

As shown in FIG. 6, the viewpoint coordinates of the image of image data set 2, that is, the viewpoint coordinates of the camera that photographed the object in the image, are extracted using the well-known technique described in Patent Document 4, for example. Then, it is saved in the support system's DB as meta information along with the image. The viewpoint coordinates (x ₀ , y ₀ , z ₀ ) of the subject are expressed, for example, in an xyz coordinate system. In this case, for example, if the subjects are vehicles, flowers, and baseball players, images of these subjects and their respective viewpoint coordinates (x ₁ , y ₁ , z ₁ ), (x ₂ , y ₂ , z ₂ ), (x ₃ , y ₃ , z ₃ ) are linked together and stored as meta information in the support system's DB for management.

When a dataset of images and viewpoints stored in the support system's database is used as a training image dataset for model learning by an image recognition engine, the distribution of viewpoints of the images used in the model is statistically analyzed and provided to the user. present. The results of the statistical analysis are displayed on the display device of the terminal device 4 (4A), for example, as a heat map of the model in spherical coordinates. In other words, when the learned model is introduced in the field, the results of statistical analysis are used to show users at the site how many images from the required viewpoints are lacking. When some images from the training image data set are used for model learning, the viewpoints of each image are counted for each viewpoint coordinate and visualized using a heat map or the like. In the example shown in Figure 6, in the heat map, areas where the counted value of the viewpoint of each image is large for each viewpoint coordinate are shown in black, areas where the counted value is medium are shown in gray, and areas where the counted value is small. is shown in white. In this example, the heat map of the model shows that there are many images taken from a viewpoint on the side of the subject, and few images taken from a viewpoint directly above the subject.

Therefore, the user can add images of missing viewpoints found from the heat map of the model by photographing them with the camera of the terminal device 4 (4A). For example, if the vehicle to be photographed is a scooter, the user uses the camera of the terminal device 4 (4A) to photograph the scooter from a viewpoint directly above and around the scooter. The terminal device 4 (4A) has a sensor that outputs sensor information used to estimate the tilt of the camera. When a user photographs a subject at the scene, focusing on the image from the missing viewpoint, the tilt of the camera estimated from the sensor information obtained by the terminal device 4 (4A) used for photographing and, for example, the center of the image The viewpoint coordinates of an image can be estimated from the positional relationship with the photographed subject. Thereby, the viewpoint coordinates of the image taken at the scene can be added to the learning image data set in the DB of the support system together with the image.

Since the user can see the heat map of the model and understand the images of the missing viewpoints, the user can take images of the subject from the missing viewpoints with the camera of the terminal device 4 (4A). Since the user only needs to take images from the missing viewpoints and add them as training image data, the system can be constructed efficiently and at low cost.

In this way, the shooting conditions of the images included in the image data set are analyzed, and images with missing shooting conditions are acquired, while new shooting conditions are also collected using the sensor of the terminal device 4 (4A) used for shooting. By doing so, data can be collected without waste and construction costs can be reduced. Further, in constructing an image recognition system, it is possible to reduce the number of steps for collecting learning image data, which requires a lot of cost (that is, human work time).

The terminal device 4 (4A) may determine whether the distribution of viewpoints is sufficient based on the presented statistical analysis results. In this case, the terminal device 4 (4A) may notify the user by displaying a message urging the user to take an image from the missing viewpoint coordinates based on the result of the statistical analysis. Since the user basically only has to perform operations according to instructions from the terminal device 4, it is convenient.

According to each of the embodiments described above, the shooting conditions of images included in the image data set are analyzed, and while images with missing shooting conditions are acquired, new shooting conditions are also acquired using the sensor of the terminal device used for shooting. collect. This makes it possible to collect data without waste and reduce the number of steps needed to build an image recognition application, thereby reducing the cost of building the image recognition application. In addition, even if the viewpoint from which the camera photographs the object cannot be fixed, for example, the learning image dataset can be analyzed and the variations in the viewpoints of the images in the training image dataset can be presented to the user. It is possible to support the collection of image data from a certain viewpoint.

Regarding the embodiments including the above examples, the following additional notes are further disclosed.
(Additional note 1)
An image recognition application construction support system capable of wireless communication with a terminal device,
a management means for storing and managing meta information indicating from which viewpoint each image of the image data set was taken in a memory together with the image;
analysis means for statistically analyzing the distribution of viewpoints of a plurality of images used by the image recognition engine for model learning among the information stored in the memory;
Presentation means for presenting the results of the statistical analysis to the terminal device;
An image recognition application construction support system characterized by:
(Additional note 2)
The presentation means makes it possible to identify images with missing viewpoints by presenting the results of the statistical analysis as a heat map of the model representing a distribution of values obtained by counting viewpoints of each image for each viewpoint coordinate. The image recognition application construction support system according to appendix 1, characterized in that the image recognition application construction support system is presented.
(Additional note 3)
An image taken by a camera of the terminal device and sensor information obtained from a sensor of the terminal device and used to estimate the tilt of the camera are received from the terminal device, and the tilt of the camera is estimated from the sensor information. further comprising estimating means for estimating a viewpoint from the positional relationship between the image and the subject in the photographed image and instructing additional storage in the memory;
The image recognition application construction according to appendix 1 or 2, wherein the management means, in response to the instruction, associates the image received by the estimation means with the estimated viewpoint and additionally stores it in the memory. support system.
(Additional note 4)
The management means is characterized in that each image of the image data set and the viewpoint coordinates of each image extracted from each image are acquired from a cloud computing system accessible from the image recognition application construction support system. , the image recognition application construction support system according to any one of Supplementary Notes 1 to 3.
(Appendix 5)
The image recognition application construction support system according to appendix 4, wherein the image data set is an open data set.
(Appendix 6)
A program that causes a computer capable of wireless communication with a terminal device to support the construction of an image recognition application,
Meta information indicating from which viewpoint each image in the image dataset was taken is stored and managed in memory along with the image.
Among the information stored in the memory, the image recognition engine performs a statistical analysis of the distribution of viewpoints of multiple images used for model learning;
presenting the results of the statistical analysis to the terminal device;
A program for causing the computer to execute processing.
(Appendix 7)
The presenting process can identify images with missing viewpoints by presenting the results of the statistical analysis as a heat map of the model that represents the distribution of values obtained by counting the viewpoints of each image for each viewpoint coordinate. The program described in Appendix 6 presented to
(Appendix 8)
An image taken by a camera of the terminal device and sensor information obtained from a sensor of the terminal device and used to estimate the tilt of the camera are received from the terminal device, and the tilt of the camera is estimated from the sensor information. and instructing additional storage in the memory by estimating a viewpoint from the positional relationship between the image and the subject in the photographed image;
causing the computer to further execute the process;
8. The program according to appendix 6 or 7, wherein the managing process associates the image received in the instructing process with the estimated viewpoint in response to the instruction, and additionally stores the linked image in the memory.
(Appendix 9)
The managing process includes any one of appendices 6 to 8, wherein each image of the image data set and the viewpoint coordinates of each image extracted from each image are acquired from a cloud computing system accessible from the computer. Programs listed in section.
(Appendix 10)
The program according to appendix 9, wherein the image data set is an open data set.

Although the disclosed image recognition application construction support system and program have been described above using examples, the present invention is not limited to the above-mentioned examples, and various modifications and improvements can be made within the scope of the present invention. Needless to say.

1, 1A Image recognition application construction support system 2 Image dataset 3 Viewpoint coordinate extraction section 4, 4A Terminal device 11 Management section 12 Analysis section 13 Presentation section 14, 14A Estimation section 100 Computer 120 Bus 121 Processor 122 Memory 123 Input device 124 Display Device 125 Interface DB Database

Claims (5)

An image recognition application construction support system capable of wireless communication with a terminal device,
a management means for storing and managing meta information indicating from which viewpoint each image of the image data set was taken in a memory together with the image;
analysis means for statistically analyzing the distribution of viewpoints of a plurality of images used by the image recognition engine for model learning among the information stored in the memory;
By presenting the results of the statistical analysis as a heat map of the model that represents the distribution of values obtained by counting the viewpoints of each image for each viewpoint coordinate, images of missing viewpoints are presented to the terminal device in such a way that they can be identified . A presentation means,
An image recognition application construction support system characterized by: An image taken by a camera of the terminal device and sensor information obtained from a sensor of the terminal device and used to estimate the tilt of the camera are received from the terminal device, and the tilt of the camera is estimated from the sensor information. further comprising estimating means for estimating a viewpoint from the positional relationship between the image and the subject in the photographed image and instructing additional storage in the memory;
The image recognition application according to claim 1 , wherein the management means additionally stores information linking the image received by the estimation means with the estimated viewpoint in the memory in response to the instruction. Construction support system. The management means is characterized in that the management means acquires each image of the image data set and the viewpoint coordinates of each image extracted from each image from a cloud computing system accessible from the image recognition application construction support system. An image recognition application construction support system according to claim 1 or 2 . 4. The image recognition application construction support system according to claim 3 , wherein the image data set is an open data set. A program that causes a computer capable of wireless communication with a terminal device to support the construction of an image recognition application,
Meta information indicating from which viewpoint each image in the image dataset was taken is stored and managed in memory along with the image.
Among the information stored in the memory, the image recognition engine performs a statistical analysis of the distribution of viewpoints of multiple images used for model learning;
By presenting the results of the statistical analysis as a heat map of the model that represents the distribution of values obtained by counting the viewpoints of each image for each viewpoint coordinate, images of missing viewpoints are presented to the terminal device in such a way that they can be identified. ,
A program for causing the computer to execute processing.

Images