JP7282122B2 - program, method, information processing device - Google Patents

Description

The present disclosure relates to programs, methods, and information processing apparatuses.

A computer game program is not only a game rule, but also a complex process that processes various elements in parallel, such as screen drawing processing, sound input/output processing, input processing from the controller, communication processing, and AI thought processing. program. Therefore, it takes a lot of man-hours to test whether each process operates normally.

JP 2019-191786 A

A test for confirming the operation of a computer game includes a test for game resources, for example, a character that appears in the game. In the test, the creator, for example, stares at the model as a game resource and confirms whether or not there is an abnormality in the model.

As a creator, I would like to increase the quality of the game by increasing the number of characters that appear in the game, but increasing the number of characters will increase the number of test cases, increasing the burden on the creator.

Patent Literature 1 describes generating a program for testing whether a game rule processing program is in a correct state. However, the patent literature does not describe testing of models of appearing characters.

The purpose of the present disclosure is to reduce the burden of testing on game resources.

A program to be executed by a computer having a processor and a memory. The program provides the processor with a step of generating a plurality of two-dimensional images based on one piece of three-dimensional model data of a game resource; and inputting each of the generated two-dimensional images to the first trained model learned as a correct output, and determining whether or not there is an abnormality in each two-dimensional image.

According to the present disclosure, it is possible to reduce the burden of testing on game resources.

1 is a block diagram showing the overall configuration of a system; FIG. 2 is a block diagram showing the configuration of the terminal device shown in FIG. 1; FIG. It is a figure which shows the functional structure of a server. FIG. 4 is a block diagram for explaining the operation when the control unit shown in FIG. 3 generates a first trained model; FIG. 4 is a block diagram for explaining the operation when the control unit shown in FIG. 3 generates a second trained model; 3 is a block diagram for explaining the operation of the control unit shown in FIG. 2 when detecting an abnormality in the three-dimensional model using the first trained model; FIG. FIG. 10 is a diagram showing an inference result presented to the user of the terminal device; 3 is a block diagram for explaining the operation of the control unit shown in FIG. 2 when detecting an abnormality in the three-dimensional model using the first trained model and the second trained model; FIG. FIG. 10 is a diagram showing an inference result presented to the user of the terminal device; FIG. 10 is a diagram showing details of an inference result presented to the user of the terminal device;

Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. In the following description, the same parts are given the same reference numerals. Their names and functions are also the same. Therefore, detailed description thereof will not be repeated.

<Overview>
The system according to the present embodiment uses the learned model to determine whether or not an abnormality has occurred in the data representing the three-dimensional model of the game resource.

<1 Configuration diagram of the entire system>
FIG. 1 is a block diagram showing an example of the overall configuration of system 1. As shown in FIG. A system 1 shown in FIG. 1 includes a plurality of terminal devices 10 and a server 20 . The terminal device 10 and the server 20 are connected for communication via a network 80, for example. The network 80 is implemented by, for example, the Internet and/or a communication network provided by a telecommunications carrier. The network 80 may be a private network to which terminals that can be connected are limited.

FIG. 1 shows an example in which the system 1 includes three terminal devices 10, but the number of terminal devices 10 included in the system 1 is not limited to three. The number of terminal devices 10 may be less than three, or may be three or more.

In this embodiment, a set of multiple devices may be used as one server. The method of distributing a plurality of functions required to realize the server 20 according to this embodiment to one or a plurality of pieces of hardware is determined in consideration of the processing capability of each piece of hardware and/or the specifications required for the server 20. can be determined as appropriate.

The terminal device 10 is, for example, a device operated by a user as a game debugger. In this embodiment, the game debugger represents, for example, someone who checks for bugs in the data representing the three-dimensional model. The user operates the terminal device 10 and confirms, for example, whether or not a bug has occurred in the data representing the three-dimensional model. The terminal device 10 is implemented by a stationary PC (Personal Computer), a laptop PC, or the like. The terminal device 10 may be realized by a mobile terminal such as a smart phone or a tablet, for example.

The terminal device 10 includes a communication IF (Interface) 12 , an input device 13 , an output device 14 , a memory 15 , a storage 16 and a processor 19 .

The communication IF 12 is an interface for inputting and outputting signals so that the terminal device 10 communicates with an external device.

The input device 13 is a device (for example, a touch panel, a touch pad, a pointing device such as a mouse, a keyboard, etc.) for receiving an input operation from a user.

The output device 14 is a device (display, speaker, etc.) for presenting information to the user.

The memory 15 temporarily stores programs and data processed by the programs, and is a volatile memory such as a DRAM (Dynamic Random Access Memory).

The storage 16 is for storing data, and is, for example, a flash memory or a HDD (Hard Disc Drive).

The processor 19 is hardware for executing an instruction set described in a program, and is composed of arithmetic units, registers, peripheral circuits, and the like.

The server 20 is, for example, a device that trains or re-learns a trained model used in the terminal device 10 . The server 20 also manages, for example, data representing a three-dimensional model as a target for removing bugs.

The server 20 includes a communication IF 22 , an input/output IF 23 , a memory 25 , a storage 26 and a processor 29 .

The communication IF 22 is an interface for inputting and outputting signals for the server 20 to communicate with an external device.

The input/output IF 23 functions as an interface with an input device for receiving input operations from the user and an output device for presenting information to the user.

The memory 25 temporarily stores programs and data processed by the programs, and is a volatile memory such as a DRAM.

The storage 26 is for storing data, and is, for example, flash memory or HDD.

The processor 29 is hardware for executing an instruction set described in a program, and is composed of arithmetic units, registers, peripheral circuits, and the like. The processor 29 includes, for example, a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit).

<1.1 Configuration of terminal device>
FIG. 2 is a block diagram showing a configuration example of the terminal device 10 shown in FIG. As shown in FIG. 2, the terminal device 10 includes a communication unit 120, an input device 13, an output device 14, an audio processing unit 17, a microphone 171, a speaker 172, a camera 160, a storage unit 180, and a control unit 190 . Each block included in the terminal device 10 is electrically connected by, for example, a bus.

The communication unit 120 performs processing such as modulation/demodulation processing for the terminal device 10 to communicate with other devices. The communication unit 120 performs transmission processing on the signal generated by the control unit 190 and transmits the signal to the outside (for example, the server 20). Communication unit 120 performs reception processing on a signal received from the outside, and outputs the signal to control unit 190 .

The input device 13 is a device for a user operating the terminal device 10 to input instructions or information. The input device 13 is realized by, for example, a mouse 131, a keyboard 132, and a touch sensitive device 133 that inputs instructions by touching an operation surface. The input device 13 converts an instruction or information input by a user into an electrical signal and outputs the electrical signal to the control unit 190 . Note that the input device 13 may include, for example, a receiving port for receiving an electrical signal input from an external input device.

The output device 14 is a device for presenting information to the user who operates the terminal device 10 . The output device 14 is implemented by, for example, the display 141 or the like. The display 141 displays data regarding accounting processing under the control of the control unit 190 . The display 141 is implemented by, for example, an LCD (Liquid Crystal Display) or an organic EL (Electro-Luminescence) display.

The audio processing unit 17 performs digital-analog conversion processing of audio signals, for example. The audio processing unit 17 converts the signal supplied from the microphone 171 into a digital signal and supplies the converted signal to the control unit 190 . Also, the audio processing unit 17 provides an audio signal to the speaker 172 . The audio processing unit 17 is realized by, for example, a processor for audio processing. The microphone 171 receives an audio input and provides an audio signal corresponding to the audio input to the audio processing section 17 . The speaker 172 converts the audio signal given from the audio processing unit 17 into audio and outputs the audio to the outside of the terminal device 10 .

The camera 160 is a device for receiving light with a light receiving element and outputting it as a photographing signal.

The storage unit 180 is realized by, for example, the memory 15 and the storage 16, and stores data and programs used by the terminal device 10. FIG. The storage unit 180 stores, for example, a first trained model 181, a second trained model 182, and three-dimensional model data 183.

The first trained model 181 and the second trained model 182 are models generated by causing the server 20 to perform machine learning on the machine learning model according to the model learning program. The first trained model 181 and the second trained model 182 are, for example, synthetic functions with parameters that are composed of multiple functions that perform predetermined inference based on input data. A parameterized composite function is defined by a combination of multiple adjustable functions and parameters. A trained model according to this embodiment may be any parameterized composite function that satisfies the above requirements.

For example, when the first trained model 181 and the second trained model 182 are generated using a forward propagation multi-layered network, the parameterized composite function is, for example, a linear relationship between each layer using a weight matrix , the non-linear relationship (or linear relationship) with the activation function in each layer, and the bias combination. Weighting matrices and biases are called parameters of the multilayered network. A parameterized composite function changes its form as a function depending on how the parameters are chosen. In a multi-layered network, by appropriately setting the constituent parameters, it is possible to define a function capable of outputting favorable results from the output layer.

As the multi-layered network according to the present embodiment, for example, a deep neural network (DNN), which is a multi-layered neural network targeted for deep learning, can be used. As the DNN, for example, a convolution neural network (CNN) targeting images may be used.

The first trained model 181 is a model that, when an image is input, outputs whether or not the image contains an abnormality. In the present embodiment, the first trained model 181 judges, for example, the following classes (types) of abnormalities in an image. It should be noted that the judgment may be made in classes other than those shown below.
・Vertex destruction (noise of polygon vertex coordinates)
・UV destruction (noise of UV coordinates of polygons)
・Dot noise ・Material abnormality (noise due to discontinuity of parameters between polygons when drawing the surface of a 3D model)
・Polygon missing ・Embedded

When an image is input, the first trained model 181 may be trained so as to indicate with a predetermined index whether any of the above abnormalities is included. For example, the predetermined index is as follows.
・Index value (probability) based on a predetermined maximum value
・Symbols representing the degree of abnormality (ABC, circle, triangle, cross)

Also, the first trained model 181 may be trained so that when an image is input, the region of interest in the image can be output in an identifiable manner. A region of interest in an image is, for example, a region in which there is a high probability that an abnormality exists.

The second trained model 182 is a model that outputs whether or not the data representing the 3D model contains bugs when judgment results of abnormalities in a plurality of images acquired for one 3D model are input. is.

The 3D model data 183 stores data representing a 3D model for each of a plurality of characters as game resources. A character as a game resource may be displayed from any direction according to the user's operation during the game. Therefore, a three-dimensional model as a game resource must be able to be viewed from all directions without anomalies.

Data representing a three-dimensional model includes, for example, polygon data and texture data. Although the data representing the three-dimensional model is not limited to polygon data and texture data, it will be explained as polygon data and texture data below.

Control unit 190 is implemented by processor 19 reading a program stored in storage unit 180 and executing instructions included in the program. The control unit 190 controls operations of the terminal device 10 . The control unit 190 functions as an operation reception unit 191 , a transmission/reception unit 192 , an image generation unit 193 , an inference unit 194 , and a display control unit 195 by operating according to programs.

The operation reception unit 191 performs processing for receiving instructions or information input from the input device 13 . Specifically, for example, the operation reception unit 191 receives information based on instructions input from the mouse 131, keyboard 132, touch-sensitive device 133, or the like.

The transmission/reception unit 192 performs processing for the terminal device 10 to transmit/receive data to/from an external device such as the server 20 according to a communication protocol.

The image generator 193 performs processing for generating a plurality of two-dimensional images from the three-dimensional model. For example, the image generation unit 193 generates two-dimensional images viewed from a plurality of viewpoints based on polygon data and texture data for one character stored in the storage unit 180 . When the three-dimensional model includes multiple frames of polygon data and texture data, the image generator 193 may generate multiple two-dimensional images from at least one frame of polygon data and texture data.

The inference unit 194 infers an output for an input using the first trained model 181 stored in the storage unit 180 .

The inference unit 194 may use the first trained model 181 and the second trained model 182 stored in the storage unit 180 to infer the output for the input. For example, the inference unit 194 infers the result of inference output using the first trained model 181 as an input to the second trained model 182 as an output.

The display control unit 195 controls the output device 14 to present the inference result to the user. Specifically, for example, the display control unit 195 causes the display 141 to display the inference result inferred by the inference unit 194 .

<1.2 Functional Configuration of Server>
FIG. 3 is a diagram showing a functional configuration of the server 20. As shown in FIG. As shown in FIG. 3 , the server 20 functions as a communication section 201 , a storage section 202 and a control section 203 .

The communication unit 201 performs processing for the server 20 to communicate with an external device.

The storage unit 202 is realized by, for example, the memory 25 and the storage 26, and stores data and programs used by the server 20. FIG. The storage unit 202 has a first trained model 2021, a second trained model 2022, and three-dimensional model data 2023, for example.

The first trained model 2021 and the second trained model 2022 have been trained on the server 20 or retrained on the server 20 . When the first trained model 2021 is updated, the updated first trained model 2021 is provided to the terminal device 10 . When the second trained model 2022 is updated, the updated second trained model 2022 is provided to the terminal device 10 .

The first trained model 2021 or the second trained model 2022 may be trained by an external device and provided to the server 20 .

The 3D model data 2023 stores polygon data and texture data for each of a plurality of characters as game resources. When the 3D model data 2023 is updated, the updated 3D model data 2023 is provided to the terminal device 10 .

The control unit 203 is implemented by the processor 29 reading a program stored in the storage unit 202 and executing instructions included in the program. The control unit 203 operates according to a program to exhibit functions shown as a reception control module 2031, a transmission control module 2032, an image generation module 2033, a processing module 2034, a learning module 2035, and an inference module 2036.

The reception control module 2031 controls processing for the server 20 to receive a signal from an external device according to a communication protocol.

The transmission control module 2032 controls the processing by which the server 20 transmits signals to external devices according to a communication protocol.

The image generation module 2033 performs processing for generating a plurality of two-dimensional images from the three-dimensional model. For example, the image generation module 2033 generates two-dimensional images viewed from multiple viewpoints based on polygon data and texture data for one character stored in the storage unit 202 . Also, for example, the image generation module 2033 generates two-dimensional images viewed from a plurality of viewpoints based on the polygon data and texture data of the three-dimensional model processed by the processing module 2034 . If the three-dimensional model includes multiple frames of polygon data and texture data, the image generation module 2033 may generate multiple two-dimensional images from each frame of polygon data and texture data.

The processing module 2034 performs processing for processing the three-dimensional model. For example, the processing module 2034 processes polygon data or texture data stored in the storage unit 202 as follows.
・Probabilistically add a noise vector to each vertex coordinate of a polygon.
Add a noise vector probabilistically to each UV coordinate of the polygon.
・Add dot noise to the texture.
- At least one of the parameters for rendering the surface of the 3D model is changed with random polygons.
・Make polygons disappear at random.
・The predetermined area (joint) is continuously displaced to generate penetration (penetration).

The learning module 2035 generates the first trained model 2021 or the second trained model 2022 by causing the machine learning model to perform machine learning based on the learning data.

Specifically, for example, the learning module 2035 uses a two-dimensional image generated from a three-dimensional model stored in the storage unit 202 as input data, and uses correct output data indicating that the two-dimensional image is normal. Train a learning model. In addition, the learning module 2035 uses the two-dimensional image generated from the three-dimensional model processed by the processing module 2034 as input data, and the information about the abnormality given by the processing of the three-dimensional model as correct output data to create a machine learning model. By learning, the first trained model 2021 is generated. The information about the abnormality given by processing the three-dimensional model includes, for example, how to give the three-dimensional model an abnormality (abnormality class), the position of the given abnormality, and the like.

Also, for example, the learning module 2035 uses the result of inference using the first trained model 2021 as input data, and corrects information about whether the 3D model is normal or abnormal given by processing the 3D model. A second trained model 2022 is generated by learning the machine learning model as output data.

The learning module 2035 receives a plurality of two-dimensional images generated from the three-dimensional model stored in the storage unit 202 as input data, and outputs a correct answer indicating that the three-dimensional model on which these two-dimensional images are based is normal. A machine learning model may be trained as data. At this time, the learning module 2035 uses, as input data, a plurality of two-dimensional images generated from the three-dimensional model processed by the processing module 2034 and given processing and processing positions, and the originals of these two-dimensional images. A second trained model 2022 is generated by making the machine learning model learn the processing and the position of the processing given to the three-dimensional model as the correct output data.

Also, the learning module 2035 re-learns the first trained model 2021 or the second trained model 2022 based on newly added learning data. Specifically, for example, the learning module 2035 re-learns the first trained model 2021 or the second trained model 2022 in the following cases.
・When a new character is added ・When a new class of error is defined

The inference module 2036 uses the first trained model 2021 stored in the storage unit 202 to infer the output for the input. The output of inference module 2036 is used as input data in training the second trained model 2022 .

<2 Operation>
(Generation of first trained model 2021)
The operation of the control unit 203 when the server 20 generates the first trained model 2021 will be described.

FIG. 4 is a block diagram for explaining the operation when the control unit 203 shown in FIG. 3 generates the first trained model 2021. As shown in FIG.

The control unit 203 reads polygon data and texture data for one character stored in the storage unit 202 . The control unit 203 uses the processing module 2034 to process the read polygon data or texture data. The processing module 2034 performs processing on, for example, read polygon data or texture data.

The control unit 203 uses the image generation module 2033 to generate two-dimensional images viewed from a plurality of viewpoints based on the polygon data and texture data processed by the processing module 2034 . The image generation module 2033 generates a plurality of two-dimensional images, for example, based on each processed polygon data and texture data.

The control unit 203 causes the machine learning model to perform machine learning using the learning module 2035 to generate the first trained model 2021 . Machine learning algorithms include discriminant analysis, logistic regression, support vector machines, neural networks, randomized trees, subspace methods, and the like.

The learning module 2035 uses, for example, the two-dimensional image generated by the image generation module 2033 as input data, and learns a machine learning model using the abnormal class, the abnormal position, etc. in the input two-dimensional image as correct output data. .

Specifically, the learning module 2035 uses as input data a two-dimensional image of a three-dimensional model in which a noise vector is added to the vertex coordinates of polygons, and determines that the abnormality class is “vertex destruction” and that the noise position is used as the correct output data to train the machine learning model.

In addition, the learning module 2035 uses as input data a two-dimensional image of a three-dimensional model in which a noise vector is added to the UV coordinates of polygons, and outputs the correct answer that the abnormality class is "UV destruction" and the position of the noise. Train a machine learning model as data.

In addition, the learning module 2035 uses a two-dimensional image of a three-dimensional model with dot noise added to the texture as input data, and the abnormality class is "dot noise" and the position of the dot noise as correct output data. Train a machine learning model.

In addition, the learning module 2035 uses as input data a two-dimensional image of a three-dimensional model in which at least one of the parameters for drawing the surface of the three-dimensional model is changed with random polygons, and the abnormality class is "material abnormality". and the position of the material anomaly are used as correct output data to learn a machine learning model.

In addition, the learning module 2035 uses as input data a two-dimensional image of a three-dimensional model in which at least one of the parameters for drawing the surface of the three-dimensional model with predetermined polygons has been changed, and the abnormality class is "material abnormality". and the position of the material anomaly are used as correct output data to learn a machine learning model.

In addition, the learning module 2035 uses a two-dimensional image of a three-dimensional model with missing polygons as input data, the abnormality class being "missing polygons", and the position of the missing polygons as correct output data. Train a machine learning model.

In addition, the learning module 2035 uses as input data a two-dimensional image of the three-dimensional model in which the embedding occurs, and the abnormality class is "embedding" and the position where the embedding occurs as correct output data. Train a learning model.

The control unit 203 uses the image generation module 2033 to generate two-dimensional images viewed from a plurality of viewpoints based on the read polygon data and texture data.

The control unit 203 uses the learning module 2035 as input data, for example, a two-dimensional image generated based on an unprocessed three-dimensional model, and determines that the input two-dimensional image is normal as correct output data. A first trained model 2021 is generated.

The control unit 203 makes the first trained model 2021 learn by repeating the above process for, for example, a predetermined character stored in the storage unit 202 . Predetermined characters are, for example, the following.
・All characters ・Characters selected by the administrator ・Characters related to people ・Characters related to creatures ・Newly designed characters (characters updated on or after a certain date)
For characters related to creatures, there are, for example, the following aspects.
・Bipedal creature ・Quadruped creature ・Flying creature ・Creature imitating an arthropod ・Creature without limbs The first trained model 2021 may be trained by repeating the combinations.

As described above, in the above-described embodiment, the control unit 203 uses the processing module 2034 to process the 3D model data of a plurality of characters so that the 3D model data of game resources includes a preset type of abnormality. . The control unit 203 uses the image generation module 2033 to generate a two-dimensional image based on the processed three-dimensional model data. The control unit 203 receives the generated two-dimensional image by the learning module 2035, and trains the first trained model 2021 with the type of abnormality given by processing as the correct output. This makes it possible to automatically create learning data for training the first trained model 2021 . For this reason, it is not necessary to store the three-dimensional model in which an abnormality has been found to create learning data for the first trained model 2021 . In addition, even when trying to generate the first trained model 2021, it is not easy to prepare a 3D model in which various error situations have occurred, and there may be cases where the number of data is small with only the actual 3D model. . In this embodiment, the data for training the first trained model 2021 is automatically created, so it is possible to obtain sufficient data for training.

In the above embodiment, the control unit 203 causes the learning module 2035 to learn the first trained model 2021 using three-dimensional model data of various types of characters as learning data. As a result, using the first trained model 2021 makes it possible to detect, with high accuracy, anomalies in the three-dimensional model data of characters that have not been trained.

In the above-described embodiment, the control unit 203 uses the processing module 2034 to add types of abnormalities such as vertex destruction, UV destruction, dot noise, material abnormalities, polygon missing, or embedding to the three-dimensional model. there is This makes it possible to automatically create learning data including anomalies that can actually occur.

Further, in the above embodiment, the control unit 203 uses the learning module 2035 to train the first trained model 2021 by using the position where the abnormality is given by processing as the correct output. As a result, an area in which an abnormality may occur is output as the output of the first trained model 2021 .

Therefore, according to the server 20 according to the present embodiment, it is possible to reduce the burden of testing the game resources.

(Generation of second trained model 2022)
The operation of the control unit 203 when the server 20 generates the first trained model 2021 will be described.

FIG. 5 is a block diagram for explaining the operation when the control unit 203 shown in FIG. 3 generates the second trained model 2022. As shown in FIG.

The control unit 203 reads polygon data and texture data for one character stored in the storage unit 202 . The control unit 203 uses the processing module 2034 to process the read polygon data or texture data.

The control unit 203 uses the image generation module 2033 to generate two-dimensional images viewed from a plurality of viewpoints based on the polygon data and texture data processed by the processing module 2034 .

The control unit 203 uses the inference module 2036 to use the first trained model 2021 stored in the storage unit 202 to infer the output for the input. Specifically, for example, the inference module 2036 inputs the two-dimensional image generated by the image generation module 2033 to the first trained model 2021 and outputs the abnormal class and abnormal position.

The control unit 203 causes the machine learning model to perform machine learning using the learning module 2035 to generate the second trained model 2022 . The learning module 2035 uses, as input data, inference results of a plurality of two-dimensional images for one three-dimensional model inferred by the inference module 2036, and the processing class and processing position given to the three-dimensional model. etc. are used as correct output data to train a machine learning model.

The control unit 203 uses the image generation module 2033 to generate two-dimensional images viewed from a plurality of viewpoints based on the read polygon data and texture data.

The control unit 203 uses the inference module 2036 to use the first trained model 2021 stored in the storage unit 202 to infer the output for the input. Specifically, for example, the inference module 2036 inputs the two-dimensional image of the unprocessed three-dimensional model generated by the image generation module 2033 to the first trained model 2021, and the two-dimensional image output the inference result about

The control unit 203 uses, as input data, inference results of a plurality of two-dimensional images of an unprocessed three-dimensional model inferred by the inference module 2036 by the learning module 2035, and determines whether the three-dimensional model is normal. A machine learning model is trained by using a certain thing as correct output data.

The control unit 203 repeats the above process for, for example, a predetermined character stored in the storage unit 202 to learn the second trained model 2022 .

As described above, in the above-described embodiment, the control unit 203 uses the learning module 2035 to input the output from the first trained model 2021 that receives a plurality of two-dimensional images, and corrects the type of abnormality given by processing. We are trying to train the second trained model 2022 as an output. When recognizing images one by one, even if the image does not contain an abnormality, even a human may misidentify an abnormality depending on how the light hits the image or the frame. Such learning data that even humans misidentify may reduce the apparent performance of AI. By using a plurality of two-dimensional images and information about anomalies in the two-dimensional images as input data, it is possible to comprehensively determine the presence or absence of anomalies in the three-dimensional model. As a result, the server 20 can improve the accuracy of detecting the presence or absence of an abnormality in the three-dimensional model.

In the above-described embodiment, the control unit 203 causes the learning module 2035 to learn the first trained model 2021 and the second trained model 2022 using three-dimensional model data of various types of characters as learning data. ing. As a result, by using the first trained model 2021 and the second trained model 2022, it is possible to detect, with high accuracy, an abnormality in the 3D model data of a character that has not been trained.

Therefore, according to the server 20 according to the present embodiment, it is possible to reduce the burden of testing the game resources.

(Inference processing 1 in terminal device 10)
The operation of the control unit 190 when the inference processing is performed in the terminal device 10 will be described.

FIG. 6 is a block diagram for explaining the operation when the control unit 190 shown in FIG. 2 uses the first trained model 181 to detect an abnormality in the three-dimensional model.

The control unit 190 reads polygon data and texture data for one character stored in the storage unit 180 . The characters of the polygon data and texture data read out at this time may be the characters used in the learning of the first trained model 181, or may be characters different from the characters used in the learning of the first trained model 181. There may be. The control unit 190 uses the image generation unit 193 to generate two-dimensional images viewed from a plurality of viewpoints based on the read polygon data and texture data. When the three-dimensional model includes multiple frames of polygon data and texture data, the image generator 193 may generate multiple two-dimensional images from at least one frame of polygon data and texture data.

The control unit 190 uses the first trained model 181 stored in the storage unit 180 by the inference unit 194 to infer the output for the input. Specifically, for example, the inference unit 194 inputs the two-dimensional image generated by the image generation unit 193 to the first abnormality determination unit 1941 realized by the first trained model 181 . The first abnormality determination unit 1941 outputs an inference result indicating whether or not the input two-dimensional image contains an abnormality. For example, the first abnormality determination unit 1941 determines an index value representing the possibility that the two-dimensional image corresponds to the normal, vertex destruction, UV destruction, dot noise, material abnormality, polygon missing, or embedded classes, and Output the region.

For example, the control unit 190 repeats the above process for the three-dimensional model of the character designated by the user.

When the inference result of the three-dimensional model of the character designated by the user is output from the inference unit 194, the control unit 190 presents the inference result to the user through the display control unit 195. FIG.

FIG. 7 is a schematic diagram showing an example of an inference result presented to the user of the terminal device 10. As shown in FIG. In FIG. 7, display 141 includes alert list 1411 and verification history list 1412 . The alert list 1411 presents two-dimensional images that are highly likely to contain anomalies. The display 141 adds to the alert list 1411, for example, a two-dimensional image whose output index value is higher than a preset threshold. The alert list 1411 includes, for example, the item "implementation date", the item "file name", the item "image number", the item "score summary", and the item "thumbnail".

The item "implementation date" represents the date on which the verification was implemented.

The item "file name" represents the file name of the verified three-dimensional model.

The item "image number" represents the identification number of the image generated for the three-dimensional model.

The item "score summary" represents a summary of the inference results output for the two-dimensional image. For example, the display control unit 195 displays, among the classes output by the inference unit 194, the class with the highest probability of being applicable in the item “score summary”. For example, the display control unit 195 displays the class with the highest index value among the classes output by the inference unit 194 in the item “score summary”.

The item "thumbnail" represents the thumbnail of the input two-dimensional image. For example, a thumbnail image of the input two-dimensional image is displayed in the area indicated by the square in FIG. When the user presses a thumbnail image, the display control unit 195 presents the user with an area of interest in the image.

A verification history list 1412 presents two-dimensional images that are verification targets. The verification history list 1412 includes, for example, the item “implementation date”, the item “file name”, the item “image number”, the item “score summary”, and the item “thumbnail”.

The user refers to the inference result displayed on the display 141 and judges an abnormality in the three-dimensional model.

The control unit 190 may transmit the inference result to another terminal device 10 . Also, the control unit 190 may transmit to the other terminal device 10 the fact that the inference has been completed instead of the inference result itself.

As described above, in the above embodiment, the control unit 190 causes the image generation unit 193 to generate a plurality of two-dimensional images based on one piece of three-dimensional model data for the game resource. The control unit 190 uses the inference unit 194 to input a two-dimensional image, and to the first trained model 181 that has been trained using information regarding the presence or absence of an abnormality in the input two-dimensional image as a correct output. By inputting each two-dimensional image, it is determined whether or not there is an abnormality in each two-dimensional image. This enables the terminal device 10 to detect the presence or absence of an abnormality in the 3D model data as game resources.

Also, one piece of 3D model data for a game resource may be different from the 3D model data used for learning the first trained model 181 . Therefore, it becomes possible to detect the presence or absence of an abnormality (bug) in the three-dimensional model data of the new character using the first trained model 181 .

Therefore, according to the terminal device 10 according to the present embodiment, it is possible to reduce the burden of testing for game resources.

In the above embodiment, the control unit 190 causes the inference unit 194 to output information representing the type of abnormality included in the input two-dimensional image to the first trained model 181 . This makes it easier for the user to confirm the abnormality.

In the above embodiment, the control unit 190 causes the inference unit 194 to store the type of anomaly included in the input two-dimensional image in the first trained model 181 and an index indicating the possibility of the type of anomaly. and is output. This makes it easier for the user to confirm the abnormality.

In the above embodiment, the control unit 190 causes the inference unit 194 to output the region to be focused on in the input two-dimensional image to the first trained model 181 . This allows the user to have an idea of the area to focus on when confirming the abnormality.

In the above embodiment, the control unit 190 causes the image generation unit 193 to generate a plurality of two-dimensional images from at least one frame of three-dimensional model data among the three-dimensional model data for a plurality of frames. . As a result, even if the three-dimensional model data stored in the storage unit 180 is a moving image, it is possible to reduce the burden on the user during the test.

(Inference process 2 in terminal device 10)
The operation of the control unit 190 when the inference processing is performed in the terminal device 10 will be described.

FIG. 8 is a block diagram for explaining the operation when the control unit 190 shown in FIG. 2 uses the first trained model 181 and the second trained model 182 to detect an abnormality in the three-dimensional model.

The control unit 190 reads polygon data and texture data for one character stored in the storage unit 180 . The characters of the polygon data and texture data read at this time may be the characters used in the learning of the first trained model 181 and the second trained model 182, or the characters used in the learning of the first trained model 181 and the second trained model. It may be a character different from the character used in learning the finished model 182 . The control unit 190 uses the image generation unit 193 to generate two-dimensional images viewed from a plurality of viewpoints based on the read polygon data and texture data. When the three-dimensional model includes multiple frames of polygon data and texture data, the image generator 193 may generate multiple two-dimensional images from at least one frame of polygon data and texture data.

The control unit 190 uses the first trained model 181 and the second trained model 182 stored in the storage unit 180 by the inference unit 194 to infer the output for the input. Specifically, for example, the inference unit 194 inputs the two-dimensional image generated by the image generation unit 193 to the first abnormality determination unit 1941 realized by the first trained model 181 . The first abnormality determination unit 1941 outputs an inference result indicating whether or not the input two-dimensional image contains an abnormality. First abnormality determination section 1941 outputs inference results for all two-dimensional images generated by image generation section 193 to second abnormality determination section 1942 realized by second trained model 182 .

The second anomaly determination unit 1942 outputs an inference result indicating whether or not an anomaly is included in the three-dimensional model from which the input inference result is based. For example, the second abnormality determination unit 1942 determines that the three-dimensional model may correspond to one of the following classes: normal, vertex destruction, UV destruction, dot noise, material abnormality, missing polygon, or embedding. An index value representing the possibility of matching or an area to be focused on is output.

For example, the control unit 190 repeats the above process for the three-dimensional model of the character designated by the user.

When the inference result of the three-dimensional model of the character designated by the user is output from the inference unit 194, the control unit 190 presents the inference result to the user through the display control unit 195. FIG.

9 and 10 are schematic diagrams showing examples of inference results presented to the user of the terminal device 10. FIG. In FIG. 9, display 141 includes alert list 1413 and verification history list 1414 . The alert list 1413 presents three-dimensional models judged to have a high possibility of containing anomalies. The display 141 adds to the alert list 1413, for example, a three-dimensional model whose output index value is higher than a preset threshold. The alert list 1413 includes, for example, the item "implementation date", the item "file name", the item "score summary", and the item "thumbnail".

The item "implementation date" represents the date on which the verification was implemented.

The item "file name" represents the file name of the verified three-dimensional model.

The item "score summary" represents a summary of the inference results output for the three-dimensional model. For example, the display control unit 195 displays, among the classes output by the inference unit 194, the class with the highest probability of being applicable in the item “score summary”. For example, the display control unit 195 displays the class with the highest index value among the classes output by the inference unit 194 in the item “score summary”.

The item "thumbnail" represents a thumbnail of a representative 2D image generated for the 3D model. A typical two-dimensional image is, for example, the following.
・2D image with the highest index value ・2D image registered in advance as a representative view

A thumbnail image of the three-dimensional model is displayed in the area indicated by the square in FIG.

A verification history list 1414 presents the three-dimensional models that have been verified. The verification history list 1414 includes, for example, the item “implementation date”, the item “file name”, the item “image number”, the item “score summary”, and the item “thumbnail”.

When the user presses one of the records displayed in the alert list 1413, the display control unit 195 causes the display 141 to display the detailed information shown in FIG. In FIG. 10, display 141 includes summary information 1415 and detail list 1416 . Summary information 1415 presents information about the record selected by the user. The summary information 1415 includes, for example, the item “implementation date”, the item “file name”, the item “score summary”, and the item “thumbnail”.

Detail list 1416 represents inference results for each 2D image generated based on the 3D model specified by the user. The detailed list 1416 includes, for example, the item "image", the item "point of interest", and the item "score".

The item “image” represents a two-dimensional image generated by the image generator 193 . This two-dimensional image may be rephrased as a two-dimensional image input to the first abnormality determination section 1941 .

The item “focus point” is a two-dimensional image output from the first abnormality determination unit 1941, and is a two-dimensional image including a target area. In this two-dimensional image, for example, an area in which an abnormality is highly likely to occur is displayed so as to be distinguished from other areas and identifiable.

The item “score” represents an index value representing the possibility of the two-dimensional image input to the first abnormality determination unit 1941 being output to the abnormality class.

The user refers to the inference result displayed on the display 141 and judges an abnormality in the three-dimensional model.

The control unit 190 may transmit the inference result to another terminal device 10 . Also, the control unit 190 may transmit to the other terminal device 10 the fact that the inference has been completed instead of the inference result itself.

As described above, in the above-described embodiment, the terminal device 10 receives, by the inference unit 194, the determination result of the presence/absence of abnormality in a plurality of two-dimensional images based on one piece of three-dimensional model data. By inputting the determination result output from the first trained model 181 to the second trained model 182 that has been trained using information regarding the presence or absence of an abnormality as a correct output, a two-dimensional image input to the first trained model 181 is obtained. It is determined whether or not there is an abnormality in the three-dimensional model data that is the basis of the . As a result, the terminal device 10 can comprehensively determine whether there is an abnormality in the three-dimensional model. Therefore, the user does not need to confirm the result of inference as to the presence or absence of abnormality in each image. That is, the terminal device 10 can reduce the burden on the user.

Also, one piece of 3D model data for a game resource may be different from the 3D model data used for learning the first trained model 181 and the second trained model 182 . Therefore, it is possible to detect whether or not there is an abnormality (bug) in the three-dimensional model data of the new character using the first trained model 181 and the second trained model 182 .

In the above embodiment, the control unit 190 causes the image generation unit 193 to generate a plurality of two-dimensional images from at least one frame of three-dimensional model data among the three-dimensional model data for a plurality of frames. The control unit 190 causes the inference unit 194 to input, to the second trained model 182, the determination result of the two-dimensional image for the three-dimensional model data of each frame, which is output from the first trained model 181. It is determined whether or not there is an abnormality in the three-dimensional model data of each frame. As a result, even if the three-dimensional model data stored in the storage unit 180 is a moving image, it is possible to reduce the burden on the user during the test.

Therefore, according to the terminal device 10 according to the present embodiment, it is possible to reduce the burden of testing for game resources.

<Modification>
In the above embodiment, the case where the image generation unit 193 or the image generation module 2033 generates two-dimensional images viewed from a plurality of viewpoints based on polygon data and texture data has been described as an example. However, the plurality of images are not limited to two-dimensional images that are viewed from different viewpoints. For example, the image generation unit 193 or the image generation module 2033 may generate multiple two-dimensional images with different lighting. Changing the lighting includes, for example, changing the position of the light source or the amount of light.

Further, in the above embodiment, the server 20 trains the first trained model 2021 or the second trained model 2022, and the trained first trained model 2021 or the second trained model 2022 is transferred to the terminal device 10. The case of providing to However, it is not limited to the server 20 that the first trained model 2021 or the second trained model 2022 is trained. The terminal device 10 may train the first trained model 2021 or the second trained model 2022 .

Although several embodiments of the present disclosure have been described above, these embodiments can be implemented in various other forms, and various omissions, replacements, and modifications are possible without departing from the spirit of the invention. It can be performed. These embodiments and their modifications are intended to be included in the scope of the invention described in the claims and their equivalents as well as included in the scope and gist of the invention.

<Appendix>
The items described in the above embodiments will be added below.

(Appendix 1)
A program to be executed by a computer comprising a processor 19 and a memory 15, wherein the program causes the processor to generate a plurality of two-dimensional images based on one three-dimensional model data 183 of the game resource ( an image generation unit 193), and a first trained model 181 that receives a two-dimensional image as an input and learns information about the presence or absence of an abnormality in the input two-dimensional image as a correct output, and a plurality of generated two-dimensional images. , respectively, to execute a step (inference unit 194) of determining whether or not there is an abnormality in each two-dimensional image.

(Appendix 2)
The program according to Appendix 1, wherein in the determining step, the first trained model outputs information representing the type of abnormality included in the input two-dimensional image.

(Appendix 3)
The program according to (Appendix 1), wherein in the determining step, the first trained model outputs the type of abnormality included in the input two-dimensional image and an index indicating the possibility of the type of abnormality.

(Appendix 4)
The program according to any one of (Appendix 1) to (Appendix 3), wherein in the determining step, the first trained model outputs a region of interest in the input two-dimensional image.

(Appendix 5)
In the judging step, the learned second model receives judgment results of the presence or absence of an abnormality in a plurality of two-dimensional images based on one piece of three-dimensional model data as an input and correct information regarding the presence or absence of an abnormality in the three-dimensional model data as a correct output. By inputting the determination result output from the first trained model to the trained model 182, it is determined whether or not there is an abnormality in the 3D model data that is the source of the 2D image input to the first trained model. The program according to any one of (Appendix 1) to (Appendix 4).

(Appendix 6)
In any of (Appendix 1) to (Appendix 5), in the step of generating a two-dimensional image, a plurality of two-dimensional images are generated from at least one frame of three-dimensional model data among a plurality of frames of three-dimensional model data. program as described.

(Appendix 7)
In the step of generating a two-dimensional image, generating a plurality of two-dimensional images from at least one frame of three-dimensional model data among a plurality of frames of three-dimensional model data, and in the determining step, the second trained model: By inputting the determination result of the two-dimensional image of the three-dimensional model data of each frame output from the first trained model, it is determined whether or not there is an abnormality in the three-dimensional model data of each frame (Appendix 5). program described in .

(Appendix 8)
A computer-implemented method comprising a processor and a memory, the method comprising: generating a plurality of two-dimensional images based on one three-dimensional model data for a game resource; , by inputting each of the generated two-dimensional images to the first trained model that has been trained using information regarding the presence or absence of an abnormality in the input two-dimensional image as a correct output, thereby determining the abnormality in each two-dimensional image. A method for performing the step of determining the presence or absence of

(Appendix 9)
An information processing apparatus 10 comprising a control unit 190 and a storage unit 180, wherein the control unit generates a plurality of two-dimensional images based on one piece of three-dimensional model data for a game resource; is input, and a plurality of generated two-dimensional images are input to the first trained model that has been trained using information regarding the presence or absence of an abnormality in the input two-dimensional image as a correct output, so that each two-dimensional image An information processing apparatus that performs a step of determining the presence or absence of an abnormality.

(Appendix 10)
A program to be executed by a computer comprising a processor 29 and a memory 25, the program instructing the processor to transform the 3D model data of the game resource so that the 3D model data of the game resource includes a preset type of anomaly. (processing module 2034), generating a two-dimensional image based on the processed three-dimensional model data (image generating module 2033), and inputting the generated two-dimensional image, processing the abnormality given by processing training the first trained model with type as correct output (learning module 2035).

(Appendix 11)
10. The program according to Appendix 10, wherein in the processing step, the type of abnormality given to the 3D model data is vertex destruction, UV destruction, dot noise, material abnormality, lack of polygon, or embedment.

(Appendix 12)
The program according to (Appendix 10) or (Appendix 11), wherein, in the training step, the first trained model is trained using the position to which the abnormality is given by processing as the correct output.

(Appendix 13)
Training a second trained model using the output from the first trained model with a plurality of two-dimensional images as input, and the type of abnormality given by processing as the correct output (Appendix 10) to (Appendix 12). Program as described in any.

(Appendix 14)
A computer-implemented method comprising a processor and a memory, wherein the processor processes three-dimensional model data for a game resource such that the three-dimensional model data includes a preset type of anomaly; A step of generating a two-dimensional image based on the processed three-dimensional model data, and a step of training a first trained model using the generated two-dimensional image as an input and correct output as the type of anomaly given by processing. how to.

(Appendix 15)
An information processing apparatus 20 comprising a control unit 203 and a storage unit 202, wherein the control unit processes the 3D model data of game resources so that the 3D model data includes a preset type of abnormality. a step of generating a two-dimensional image based on the processed three-dimensional model data; a step of training a first trained model using the generated two-dimensional image as an input and correct output as the type of anomaly given by processing. and an information processing device that executes

1... System 10... Terminal device 12... Communication IF
120... Communication unit 13... Input device 131... Mouse 132... Keyboard 133... Touch sensitive device 14... Output device 141... Display 1411... Alert list 1412... Verification history list 1413... Alert list 1414... Verification history list 1415... Summary information 1416 Detailed list 15 Memory 16 Storage 160 Camera 17 Sound processing unit 171 Microphone 172 Speaker 180 Storage unit 181 First trained model 182 Second trained model 183 Three-dimensional model data 19 Processor 190 Control unit 191 Operation receiving unit 192 Transmission/reception unit 193 Image generation unit 194 Inference unit 1941 First abnormality determination unit 1942 Second abnormality determination unit 195 Display control unit 20 Server 201 Communication unit 202 Storage unit 2021 First trained model 2022 Second trained model 2023 Three-dimensional model data 203 Control unit 2031 Reception control module 2032 Transmission control module 2033 Image generation module 2034 Processing module 2035 Learning module 2036 -- Inference module 22 -- Communication IF
23 input/output IF
25 Memory 26 Storage 29 Processor 80 Network

Claims (15)

A program to be executed by a computer comprising a processor and a memory, the program comprising:
generating a plurality of two-dimensional images based on one three-dimensional model data for the game resource;
By inputting each of the generated two-dimensional images into a first trained model trained with a two-dimensional image as an input and information regarding the presence or absence of an abnormality in the input two-dimensional image as a correct output, each A program for executing a step of determining whether or not there is an abnormality in a two-dimensional image. 2. The program according to claim 1, wherein in said determining step, said first trained model outputs information representing a type of abnormality contained in an input two-dimensional image. 2. The program according to claim 1, wherein in said determining step, said first trained model outputs a type of anomaly included in an input two-dimensional image and an index indicating the possibility of said type of anomaly. 4. The program according to any one of claims 1 to 3, wherein in said determining step, said first trained model outputs a region to be focused on in an input two-dimensional image. In the judging step, learning is performed with judgment results of the presence or absence of anomalies in a plurality of two-dimensional images based on one piece of three-dimensional model data as inputs, and information on the presence or absence of anomalies in the three-dimensional model data as a correct output. By inputting the determination result output from the first trained model to the second trained model, the presence or absence of abnormality in the three-dimensional model data that is the basis of the two-dimensional image input to the first trained model is checked. 5. The program according to any one of claims 1 to 4, for determining. 6. The program according to any one of claims 1 to 5, wherein in the step of generating the two-dimensional image, a plurality of two-dimensional images are generated from at least one frame of three-dimensional model data out of a plurality of frames of three-dimensional model data. . generating a plurality of two-dimensional images from at least one frame of three-dimensional model data among a plurality of frames of three-dimensional model data, in the step of generating two-dimensional images;
In the determination step, by inputting the determination result of the two-dimensional image of the three-dimensional model data of each frame, which is output from the first trained model, to the second trained model, 6. The program according to claim 5, which determines whether or not there is an abnormality in the three-dimensional model data. A computer-implemented method comprising a processor and a memory, wherein the processor comprises:
generating a plurality of two-dimensional images based on one three-dimensional model data for the game resource;
By inputting each of the generated two-dimensional images into a first trained model trained with a two-dimensional image as an input and information regarding the presence or absence of an abnormality in the input two-dimensional image as a correct output, each and determining whether there is an abnormality in the two-dimensional image. An information processing device comprising a control unit and a storage unit, wherein the control unit
generating a plurality of two-dimensional images based on one three-dimensional model data for the game resource;
By inputting each of the generated two-dimensional images into a first trained model trained with a two-dimensional image as an input and information regarding the presence or absence of an abnormality in the input two-dimensional image as a correct output, each and determining whether or not there is an abnormality in the two-dimensional image. A program to be executed by a computer comprising a processor and a memory, the program comprising:
processing the three-dimensional model data for the game resource so that the three-dimensional model data includes a preset type of anomaly;
generating a two-dimensional image based on the processed three-dimensional model data;
a step of training a first trained model using the generated two-dimensional image as an input and correct output as the type of anomaly given by processing. 11. The program according to claim 10, wherein in said processing step, the type of abnormality given to said three-dimensional model data is vertex destruction, UV destruction, dot noise, material abnormality, lack of polygon, or embedment. 12. The program according to claim 10, wherein, in said training step, said first learned model is trained by using a position to which an abnormality is given by said processing as a correct output. 13. Any one of claims 10 to 12, wherein a second trained model is trained with an output from a first trained model that receives a plurality of two-dimensional images as an input, and a type of abnormality given by said processing as a correct output. program as described. A computer-implemented method comprising a processor and a memory, wherein the processor comprises:
processing the three-dimensional model data for the game resource so that the three-dimensional model data includes a preset type of anomaly;
generating a two-dimensional image based on the processed three-dimensional model data;
and training a first trained model using the generated two-dimensional image as an input and the type of anomaly given by processing as a correct output. An information processing device comprising a control unit and a storage unit, wherein the control unit
processing the three-dimensional model data for the game resource so that the three-dimensional model data includes a preset type of anomaly;
generating a two-dimensional image based on the processed three-dimensional model data;
and training a first learned model using the generated two-dimensional image as an input and the type of anomaly given by processing as a correct output.

Images