Discrimination Method, Device, Server and Storage Medium for Abnormal Model Drawing

Abstract

Embodiments of the present application disclose a method, apparatus, server, and storage medium for discriminating abnormal rendering of a model; embodiments of the present application can obtain the design drawing of a target model, where the target model is a game virtual model; based on a preset first angle or a preset second angle, perform rendering processing on the target model in the game to obtain a rendering drawing of the target model; obtain the visual similarity between the design drawing and the rendering drawing; when the visual similarity is within a preset range, determine the discrimination result as that the target model is normally rendered in the game; when the visual similarity is not within the preset range, determine the discrimination result as that the target model is abnormally rendered in the game; and display the discrimination result of the target model to the user. In the embodiments of the present application, the design drawing of the game virtual model can be automatically checked against its rendering effect in the game. Thus, this solution can improve the efficiency of image discrimination.

Description

Technical Field

[0001] The present application relates to the field of computers, and in particular to a method, device, server, and storage medium for distinguishing abnormal model drawing. Background Art

[0002] Game original paintings are an important part of the early stage of game production. The original artist designs the art plan of the entire game based on the planning copy, such as the appearance design drawings of models, characters, monsters, etc., to provide standards and basis for the production of later game models.

[0003] However, in the current game production process, due to the large scale and frequent changes of game models, the final drawings drawn in the game may be quite different from the originally designed appearance design drawings. Therefore, it is often discovered that the game models in the game screen do not match their original appearance design drawings after the game is released.

[0004] Therefore, there is an urgent need for an efficient and accurate method to identify abnormal model drawing to determine the similarity between the design drawing and the drawing, so as to discover and solve the problem in time. Summary of the Invention

[0005] The embodiments of the present application provide a method, device, server and storage medium for determining abnormal model drawing, which can efficiently determine the similarity between the appearance design drawing and the drawing, and improve the accuracy of image recognition.

[0006] The present invention provides a method for determining abnormal model drawing, including:

[0007] Obtaining an appearance design drawing of a target model, where the target model is a game model;

[0008] Based on the preset first angle or the preset second angle, the target model is drawn in the game to obtain a drawing of the target model;

[0009] Obtaining the visual similarity between the design drawing and the rendered drawing;

[0010] When the visual similarity is within the preset range, the judgment result is determined to be that the target model is drawn normally in the game;

[0011] When the visual similarity is not within the preset range, the discrimination result is determined as the target model being drawn abnormally in the game;

[0012] Display the discrimination results of the target model to the user.

[0013] The present application also provides a device for determining abnormal model drawing, including:

[0014] An acquisition unit for acquiring the design drawing of a target model, where the target model is a game model;

[0015] A drawing unit for performing drawing processing on the target model in the game based on a preset first angle or a preset second angle to obtain a drawing of the target model;

[0016] A similarity unit for obtaining the visual similarity between the design drawing and the drawing;

[0017] A normal unit for, when the visual similarity is within a preset range, determining the discrimination result as that the target model is drawn normally in the game;

[0018] An abnormal unit for, when the visual similarity is not within the preset range, determining the discrimination result as that the target model is drawn abnormally in the game;

[0019] A display unit for displaying the discrimination result of the target model to the user.

[0020] In some embodiments, the similarity unit includes:

[0021] A design drawing subunit for acquiring the design drawing of the game model;

[0022] A first drawing subunit for performing drawing processing on the game model in the game based on a preset first angle to obtain a first drawing of the game model;

[0023] A second drawing subunit for performing drawing processing on the game model in the game based on a preset second angle to obtain a second drawing of the game model;

[0024] A data subunit for obtaining training data based on the design drawing, the first drawing, and the second drawing of the game model;

[0025] A training subunit for training a preset model with the training data until the preset model converges to obtain a feature extraction network;

[0026] A similarity subunit for determining the visual similarity between the design drawing and the drawing by using the feature extraction network.

[0027] In some embodiments, the game model includes a model to be processed and other models, where the other models are other game models except the model to be processed. The data subunit includes:

[0028] An anchor point submodule for adding anchor point labels to the design drawing of the model to be processed to obtain anchor point samples;

[0029] A positive sample submodule for adding positive labels to the first drawing or the second drawing of the model to be processed to obtain positive samples;

[0030] The negative sample sub-module is used to add negative labels to the first drawing or the second drawing of other models to obtain negative samples;

[0031] The data sub-module is used to obtain training data, and the training data includes anchor samples, positive samples, and negative samples.

[0032] In some embodiments, the preset model includes a first preset network, a second preset network, and a third preset network. The first preset network, the second preset network, and the third preset network share the same network parameters. The training subunit includes:

[0033] The anchor vector sub-module is used to extract the feature vector of the anchor sample by using the first preset network;

[0034] The positive sample vector sub-module is used to extract the feature vector of the positive sample by using the second preset network;

[0035] The negative sample vector sub-module is used to extract the feature vector of the negative sample by using the third preset network;

[0036] The update sub-module is used to update the network parameters of the first preset network, the second preset network, and the third preset network according to the feature vector of the anchor sample, the feature vector of the positive sample, and the feature vector of the negative sample until the preset model converges, and a feature extraction network with the same three network parameters is obtained.

[0037] In some embodiments, the update sub-module is used to:

[0038] Calculate the vector distance between the feature vector of the anchor sample and the feature vector of the positive sample to obtain a first distance;

[0039] Calculate the vector distance between the feature vector of the anchor sample and the feature vector of the negative sample to obtain a second distance;

[0040] Calculate the loss value, and the loss value is the sum of the distance difference and the preset threshold. The distance difference is the difference between the first distance and the second distance;

[0041] Use the loss value to update the network parameters of the first preset network, the second preset network, and the third preset network until the loss value is not greater than 0, and a feature extraction network with the same three network parameters is obtained.

[0042] In some embodiments, the similarity unit includes:

[0043] The construction subunit is used to construct a similarity calculation model, and the similarity calculation model includes two feature extraction networks with the same network parameters;

[0044] The design drawing vector subunit is used to calculate the feature vector of the appearance design drawing by using one of the feature extraction networks in the similarity calculation model;

[0045] A drawing graph quantum unit for calculating the feature vector of a drawing graph by using another feature extraction network in a similarity calculation model;

[0046] A distance unit for using the vector distance between the feature vector of a design drawing and the feature vector of a drawing graph as the visual similarity between the design drawing and the drawing graph.

[0047] In some embodiments, the feature extraction network includes a convolutional layer, the convolutional layer includes a plurality of residual blocks, and the design drawing vector unit is configured to:

[0048] Determine the output result of the previous residual block and the current residual block located after the previous residual block;

[0049] Perform convolutional processing on the output result of the previous residual block in the current residual block to obtain a convolutional result;

[0050] Add the convolutional result and the output result of the previous residual block to obtain the output result of the current residual block;

[0051] Calculate the feature vector of the design drawing based on the convolutional matrix, where the convolutional matrix is the output result of the last residual block.

[0052] In some embodiments, the feature extraction network includes a convolutional layer, the convolutional layer includes depthwise convolution, and the design drawing vector unit includes:

[0053] A mapping sub-module for obtaining a preset mapping relationship, where the mapping relationship is used to describe the depthwise convolution corresponding to each channel, and the design drawing has multiple channels;

[0054] A depthwise convolution sub-module for performing convolutional processing on the design drawing of the channel by using the depthwise convolution corresponding to the channel to obtain the depthwise convolution result of the channel;

[0055] An integration sub-module for integrating the depthwise convolution results of all channels into a convolutional matrix;

[0056] A design drawing vector sub-module for calculating the feature vector of the design drawing based on the convolutional matrix.

[0057] In some embodiments, the integration sub-module is configured to:

[0058] Perform slicing processing on the depthwise convolution result of the channel to obtain multiple result segments of the channel;

[0059] Perform shuffling processing on the multiple result segments of all channels to obtain shuffled result segments;

[0060] Integrate the shuffled result segments into a convolutional matrix;

[0061] Calculate the feature vector of the design drawing based on the convolutional matrix.

[0062] In some embodiments, the convolutional matrix includes a plurality of matrix elements, and the feature extraction network further includes a weight ignoring layer, a hidden layer, a normalization layer, and an embedding layer. When the design drawing vector subunit is used to calculate the feature vector of the design drawing based on the convolutional matrix, it is specifically used for:

[0063] In the weight ignoring layer, randomly select matrix elements from the convolutional matrix as elements to be set to zero, and perform zeroing processing on the elements to be set to zero in the convolutional matrix to obtain a processed convolutional matrix;

[0064] In the hidden layer, perform dimension conversion processing on the processed convolutional matrix to obtain a hidden layer result;

[0065] In the normalization layer, perform normalization processing on the hidden layer result to obtain a normalized result;

[0066] In the embedding layer, perform vector embedding processing on the normalized result to obtain the feature vector of the design drawing.

[0067] In some embodiments, the drawing unit is used for:

[0068] Obtain a fitting loading instruction;

[0069] Search for a target fitting in the game model library according to the fitting loading instruction;

[0070] Load the target fitting on the target model to obtain a loaded target model;

[0071] Based on a preset first angle or a preset second angle, perform drawing processing on the loaded target model in the game to obtain a drawing of the target model.

[0072] In some embodiments, the obtaining unit is used for:

[0073] Obtain a list of model names, where the list of model names includes the model names of game models;

[0074] Based on the model name of the target model, search for the design drawing of the target model in the design drawing database;

[0075] Show the discrimination result of the target model to the user, including:

[0076] Send the design drawing, the drawing, and the discrimination result of the target model to the client, so that the client can show the discrimination result of the target model to the user.

[0077] An embodiment of the present application further provides a server, including a memory storing multiple instructions; the processor loads the instructions from the memory to execute the steps in any of the discriminant methods for abnormal drawing of models provided by the embodiments of the present application.

[0078] An embodiment of the present application further provides a computer-readable storage medium storing multiple instructions, which are suitable for being loaded by a processor to execute the steps in any of the discriminant methods for abnormal drawing of models provided by the embodiments of the present application.

[0079] An embodiment of the present application can obtain the design drawing of a target model, where the target model is a game model; perform drawing processing on the target model in the game based on a preset first angle or a preset second angle to obtain a drawing of the target model; obtain the visual similarity between the design drawing and the drawing; when the visual similarity is within a preset range, the design drawing and the drawing are similar, so the discriminant result is determined to be that the target model is drawn normally in the game; when the visual similarity is not within the preset range, the design drawing and the drawing are not similar, so the discriminant result is determined to be that the target model is drawn abnormally in the game; display the discriminant result of the target model to the user.

[0080] In the present application, the game model can be automatically drawn at different angles in the game, and the drawing result is compared with the design drawing of the game model, so as to determine whether the design drawing and the drawing are similar, and finally the discriminant result is displayed to the user. Thus, the present application can efficiently determine the similarity between the design drawing and the drawing, and improve the accuracy of image discrimination. Description of the Drawings

[0081] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those skilled in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0082] Figure 1a It is a schematic diagram of the scenario of the discriminant method for abnormal drawing of models provided by the embodiments of the present application;

[0083] Figure 1b It is a schematic flowchart of the discriminant method for abnormal drawing of models provided by the embodiments of the present application;

[0084] Figure 1c It is a schematic diagram of the prop rotation of the discriminant method for abnormal drawing of models provided by the embodiments of the present application;

[0085] Figure 1dIt is a schematic diagram of accessory assembly for the discriminant method of model abnormal drawing provided by an embodiment of the present application;

[0086] Figure 1e It is a schematic diagram of sample annotation for the discriminant method of model abnormal drawing provided by an embodiment of the present application;

[0087] Figure 1f It is a schematic diagram of the training process for the discriminant method of model abnormal drawing provided by an embodiment of the present application;

[0088] Figure 1g It is a schematic diagram of vector distance for the discriminant method of model abnormal drawing provided by an embodiment of the present application;

[0089] Figure 1h It is a schematic diagram of the similarity calculation model structure for the discriminant method of model abnormal drawing provided by an embodiment of the present application;

[0090] Figure 2a It is a schematic diagram of the process for the discriminant method of model abnormal drawing applied in the game test scenario provided by an embodiment of the present application;

[0091] Figure 2b It is a schematic diagram of the appearance design drawing and the drawing in the firearm preview interface for the discriminant method of model abnormal drawing provided by an embodiment of the present application;

[0092] Figure 2c It is a schematic diagram of firearm painting for the discriminant method of model abnormal drawing provided by an embodiment of the present application;

[0093] Figure 2d It is a schematic diagram of the Loss curve for the discriminant method of model abnormal drawing provided by an embodiment of the present application;

[0094] Figure 2e It is a schematic diagram of the residual block structure for the discriminant method of model abnormal drawing provided by an embodiment of the present application;

[0095] Figure 2f It is a schematic diagram of the principle of depthwise separable convolution for the discriminant method of model abnormal drawing provided by an embodiment of the present application;

[0096] Figure 2g It is a schematic diagram of the feature extraction network structure for the discriminant method of model abnormal drawing provided by an embodiment of the present application;

[0097] Figure 2h It is a schematic diagram of displaying the discriminant result for the discriminant method of model abnormal drawing provided by an embodiment of the present application;

[0098] Figure 2i It is a schematic diagram of the ROC curve for the discriminant method of model abnormal drawing provided by an embodiment of the present application;

[0099] Figure 3 It is a schematic structural diagram of a discriminant device for abnormal model drawing provided by an embodiment of the present application;

[0100] Figure 4 It is a schematic structural diagram of a server provided by an embodiment of the present application. Specific embodiments

[0101] Next, the technical solutions in the embodiments of the present application will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0102] An embodiment of the present application provides a method, device, server, and storage medium for discriminating abnormal model drawing.

[0103] Among them, the discriminant device for abnormal model drawing can be specifically integrated in an electronic device, and the electronic device can be a device such as a terminal or a server. Among them, the terminal can be a device such as a mobile phone, a tablet computer, a smart Bluetooth device, a laptop computer, or a personal computer (PC); the server can be a single server or a server cluster composed of multiple servers.

[0104] In some embodiments, the discriminant device for abnormal model drawing can also be integrated in multiple electronic devices. For example, the discriminant device for abnormal model drawing can be integrated in multiple servers, and the discriminant method for abnormal model drawing of the present application is implemented by multiple servers.

[0105] In some embodiments, the server can also be implemented in the form of a terminal.

[0106] For example, referring to Figure 1a , the electronic device can be a server, and the server can obtain the appearance design drawing of the target model, where the target model is a game model; then, based on a preset first angle or a preset second angle, perform drawing processing on the target model in the game to obtain a drawing of the target model; obtain the visual similarity between the appearance design drawing and the drawing; when the visual similarity is within a preset range, determine that the discriminant result is that the appearance design drawing and the drawing are similar; when the visual similarity is not within the preset range, determine that the discriminant result is that the appearance design drawing and the drawing are not similar; finally, display the discriminant result of the target model to the user, for example, send the discriminant result back to the client, and the client displays the discriminant result to the user.

[0107] The following will be described in detail respectively. It should be noted that the serial numbers of the following embodiments are not used to limit the preferred order of the embodiments.

[0108] Artificial Intelligence (AI) is a technology that uses digital computers to simulate human perception of the environment, acquisition of knowledge, and use of knowledge. This technology enables machines to have functions similar to human perception, reasoning, and decision-making. Artificial intelligence technology mainly includes several aspects such as computer vision technology, speech processing technology, natural language processing technology, as well as machine learning and deep learning.

[0109] Among them, Computer Vision (CV) is a technology that uses a computer to replace the human eye to perform operations such as recognition and measurement on a target image and further process it. Computer vision technology usually includes technologies such as image processing, image recognition, image semantic understanding, image retrieval, virtual reality, augmented reality, simultaneous localization and mapping, etc. For example, image processing technologies such as image coloring and image stroke extraction.

[0110] In this embodiment, a method for discriminating abnormal drawing of a model based on computer vision is provided. As Figure 1b shown, the specific process of this method for discriminating abnormal drawing of the model can be as follows:

[0111] 101. Obtain the design drawing of the appearance of the target model, where the target model is a game model.

[0112] A game model is a virtual model in a game, including people, objects, buildings, particle effects, etc., such as game character models, game building models, game terrain models, game prop models, etc.

[0113] Among them, game character models can include models of humans, animals, fantasy creatures, plants, etc.; game building models can include models of houses, villages, churches, streets, doors, walls, etc.; game terrain models can include models of mountains, forests, basins, rocks, streams, etc.

[0114] Among them, game prop models can include models of game equipment, consumables, mission models, etc. Game equipment can include models of game weapons such as guns, swords, staffs, bows, etc. Game equipment can also include models of game clothing and protective gear such as belts, shoes, helmets, and armors; Consumables can include models of forging materials, food, grenades, mines, etc.; Mission models can include models of letters, keys, etc.

[0115] The design drawing of the appearance of the game model is an art plan designed for the game model in the early stage of game production. This art plan can be presented in the form of a two-dimensional image to provide standards and basis for the production of later game models, etc. The game model can be drawn in the game screen through computer drawing, and this game screen is the drawing of the game model.

[0116] This application can automatically and efficiently determine the visual similarity between the original design drawing of a game model and the drawing in the final game, and display the design drawing, the drawing, and the determination result to the user, improving the user experience.

[0117] There are various methods to obtain the design drawing of a target model. For example, in some embodiments, the design drawing of the target model can be obtained from a design drawing database through the network; for example, in some embodiments, the design drawing of the target model sent by a client can be obtained through the network, and so on.

[0118] For example, in some embodiments, the client can send the model name of the game model to be determined to the server, and the server automatically determines the visual similarity between the design drawing and the drawing of these game models in sequence, and then displays the design drawing, the drawing, and the determination result of these game models to the user.

[0119] In some embodiments, the client can send the model names of the game models to be determined to the server in the form of a list, and the server automatically determines the visual similarity between the design drawing and the drawing of these game models in the list in sequence, and then displays the design drawing, the drawing, and the determination result of these game models to the user.

[0120] Therefore, in some embodiments, step 101 may include the following steps:

[0121] Obtain a list of model names, which may include the model names of game models;

[0122] Based on the model name of the target model, search for the design drawing of the target model in the design drawing database;

[0123] In step 105, the design drawing, the drawing, and the determination result of the target model can be sent to the client so that the client can display the determination result to the user.

[0124] Among them, the determination result may include the normal drawing of the target model in the game and the abnormal drawing of the target model in the game. For example, when the design drawing and the drawing are similar, it means that the target model is drawn normally in the game, and when the design drawing and the drawing are not similar, it means that there is a problem with the abnormal drawing of the target model in the game. The similarity between the design drawing and the drawing will be introduced in detail in step 103 and will not be elaborated here.

[0125] It should be noted that, in some embodiments, the model name can also be represented in the form of a number, that is, the list of model names may include the model numbers of game models.

[0126] In some embodiments, the model name list may include both the model name and model number of the game model at the same time.

[0127] For example, in some embodiments, the game model may be a game weapon. Referring to Table 1, the model name list may include the weapon number and weapon name of the game weapon:

[0128] Table 1

[0129] Number Name 000 Semiautomatic Pistol A 001 Bolt-Action Sniper Rifle B 002 Short-Barreled Shotgun C

[0130] After the server obtains the model name list, it can search the design drawing database for the appearance design drawings of the semi-automatic pistol A with the number 000, the appearance design drawings of the bolt-action sniper rifle B with the number 001, and the appearance design drawings of the short-barreled shotgun C with the number 002.

[0131] In step 105, the server can send the appearance design drawings, drawing diagrams, and discrimination results of the automatic pistol A, the appearance design drawings, drawing diagrams, and discrimination results of the bolt-action sniper rifle B, and the appearance design drawings, drawing diagrams, and discrimination results of the short-barreled shotgun C to the client, so that the client can display the discrimination results to the user.

[0132] 102. Based on a preset first angle or a preset second angle, perform a drawing process on the target model in the game to obtain a drawing diagram of the target model.

[0133] Among them, the preset first angle and the preset second angle are preset angles, and the game model can rotate around a preset rotation point or rotation axis in the game based on this angle.

[0134] Since each game model has an initial placement angle in the game, therefore, in step 102, it can rotate around a preset rotation point or rotation axis in the game based on this angle on the basis of the initial placement angle. This rotation point or rotation axis can be located inside the game model, on the surface of the game model, or outside the game model.

[0135] In some embodiments, the game model can be rotated by a preset first angle or a preset second angle, so as to perform a drawing process on the rotated target model to obtain a drawing diagram of the target model.

[0136] In some embodiments, the game model can be rotated to a preset first angle or a preset second angle, so as to perform a drawing process on the rotated target model to obtain a drawing diagram of the target model.

[0137] It should be noted that, in addition to the preset first angle and the preset second angle, in some embodiments, based on multiple angles such as the preset third angle, the preset fourth angle, the preset fifth angle, etc., the target model can be drawn in the game to obtain the drawing of the target model.

[0138] In some embodiments, the difference between the preset first angle and the preset second angle can be 180°; for example, in some embodiments, the preset first angle can be 0°, and the preset second angle can be 180°; for example, in some embodiments, the preset first angle can be 90°, and the preset second angle can be 270°, and so on.

[0139] In some embodiments, the central axis of the game model can also be the axis of symmetry. The left side surface of the game model on the left side of the axis of symmetry and the right side surface on the right side of the axis of symmetry are symmetrical to each other. For example, for game guns, the left side surface and the right side surface of some game guns are mirror-symmetrical, and the left side surface and the right side surface of the game guns can well display the shape, color and other features of the guns. Therefore, the game images of the left side surface and the right side surface of these game guns can both be used as the drawing.

[0140] Since the left side surface and the right side surface of these game guns are the same in visual effect, therefore, regardless of whether the rotation angle is the preset first angle or the preset second angle, the final obtained drawing is the drawing of the game model.

[0141] For example, referring to Figure 1c , in some embodiments, the game model is a game gun, the preset first angle is 0°, the preset second angle is 180°, and in the gun display scene of the game, the side surface of the gun body of the game gun is perpendicular to the player's perspective. In step 102, the game gun can be rotated counterclockwise by 0° or 180° around the z-axis, so as to perform drawing processing on the rotated gun model to obtain the drawing of the target model.

[0142] Accessories can be installed on the game models in the game. Accessories refer to virtual models such as parts and components assembled on the game models. Accessories can include weapon accessories, clothing accessories, character accessories, etc.; for example, weapon accessories can include weapon ornaments, weapon special effect accessories, gun sights, gun grips, etc.; clothing accessories can include corsets, belts, shawls, capes, pendants, etc.; character accessories can include hairstyles, beards, glasses, necklaces, earrings, etc.

[0143] For example, when the game model is a gun model, referring to Figure 1d , in the game, various sight accessories, muzzle accessories, grip accessories, etc. can be installed on the game gun.

[0144] Therefore, in some embodiments, when rendering the target model in the game, the accessories loaded on the target model can also be rendered. Step 102 may include the following steps:

[0145] Obtain an accessory loading instruction;

[0146] Search for the target accessory in the game model library according to the accessory loading instruction;

[0147] Load the target accessory onto the target model to obtain the loaded target model;

[0148] Based on a preset first angle or a preset second angle, perform rendering processing on the loaded target model in the game to obtain a rendering of the target model.

[0149] Wherein, the target accessory is an accessory model in the game.

[0150] Wherein, the accessory loading instruction can carry information such as the number and name of the accessory, and is used to instruct the discriminator for abnormal model rendering to find a specific accessory and assemble it on the target model.

[0151] In some embodiments, the accessory loading instruction issued by the user can be obtained from the client. That is, the user can select the target accessory from multiple game accessories on the client, and the client loads information such as the name and number of the target accessory on the accessory loading instruction, and sends the accessory loading instruction to the discriminator for abnormal model rendering.

[0152] It should be noted that the rendering of the target model in the game can be completed by the client or by the server. That is, the game program can be loaded and run on the client, and the client can render the target model in the game; the game program can also be loaded and run on the server, and the server can render the target model in the game.

[0153] 103. Obtain the visual similarity between the design drawing and the rendering.

[0154] Visual similarity is a parameter used to characterize the degree of visual similarity, such as the similarity degree in many visual dimensions such as color, shape, and texture.

[0155] In the embodiments of the present application, the visual similarity between the design drawing and the rendering can be calculated through traditional image processing techniques, or the visual similarity between the design drawing and the rendering can be calculated through CV techniques.

[0156] For example, in some embodiments, traditional image processing techniques can be used to calculate the visual similarity between the design drawing and the rendering. For example, the visual similarity is determined by comparing the differences in traditional image features between the design drawing and the rendering.

[0157] For example, in some embodiments, a feature extraction network may be used to determine the visual similarity between a design drawing and a rendering drawing. The feature extraction network may be an artificial neural network (ANNs). The feature extraction network may communicate with a network server to obtain data therefrom, or may be trained by other artificial neural networks.

[0158] For example, in some embodiments, the feature extraction network may be trained through model training. Thus, step 103 may include the following steps:

[0159] (1) Obtain the design drawing of the game model;

[0160] (2) Perform rendering processing on the game model in the game based on a preset first angle to obtain a first rendering drawing of the game model;

[0161] (3) Perform rendering processing on the game model in the game based on a preset second angle to obtain a second rendering drawing of the game model;

[0162] (4) Obtain training data based on the design drawing, the first rendering drawing, and the second rendering drawing of the game model;

[0163] (5) Train a preset model using the training data until the preset model converges to obtain the feature extraction network;

[0164] (6) Use the feature extraction network to determine the visual similarity between the design drawing and the rendering drawing.

[0165] The first rendering drawing is a game screen of the game model rendered in the game based on the preset first angle, and the second rendering drawing is a game screen of the game model rendered in the game based on the preset second angle.

[0166] In some embodiments, the game model may be a game firearm, the preset first angle may be 0°, and the preset second angle may be 180°. In the firearm display scene of the game, the side of the gun body of the game firearm may be perpendicular to the player's perspective. In step (1) "Perform rendering processing on the game model in the game based on the preset first angle to obtain a first rendering drawing of the game model", the game firearm may be rotated counterclockwise by 0° around the z-axis, and then the rotated firearm model may be rendered to obtain the first rendering drawing of the game firearm. Then, in step (2) "Perform rendering processing on the game model in the game based on the preset second angle to obtain a second rendering drawing of the game model", the game firearm may be rotated counterclockwise by 180° around the z-axis, and then the rotated firearm model may be rendered to obtain the second rendering drawing of the game firearm.

[0167] In some embodiments, the differences between game models are small. Therefore, the appearance design drawings, the first drawing, and the second drawing of the game models can all be used as training data to train a preset model, so that the finally trained feature extraction network can more accurately distinguish the game models.

[0168] For example, in some embodiments, the game model may include a model to be processed and other models. The other models are other game models except the model to be processed. Step (4) "obtaining training data based on the appearance design drawing, the first drawing, and the second drawing of the game model" may include the following steps:

[0169] Adding anchor point labels to the appearance design drawing of the model to be processed to obtain anchor point samples;

[0170] Adding positive labels to the first drawing or the second drawing of the model to be processed to obtain positive samples;

[0171] Adding negative labels to the first drawing or the second drawing of the other models to obtain negative samples;

[0172] Obtaining training data, which may include anchor point samples, positive samples, and negative samples.

[0173] For example, referring to Figure 1e , the game model may be a game firearm, including rifle 001, rifle 002, and rifle 003; when the model to be processed is rifle 001, the appearance design drawing of rifle 001 can be added with anchor point labels as anchor point samples; the first drawing of rifle 001 can be added with positive labels as positive sample 1, or the second drawing of rifle 001 can be added with positive labels as positive sample 2; the first drawing of rifle 002 can be added with negative labels as negative sample 1, or the second drawing of rifle 002 can be added with negative labels as negative sample 2; the first drawing of rifle 003 can be added with negative labels as negative sample 3, or the second drawing of rifle 003 can be added with negative labels as negative sample 4.

[0174] In the embodiments of the present application, an anchor point sample, a positive sample, and a negative sample can be used to train a preset model. The preset model can be used to extract image features, and the preset model can be a Convolutional Neural Networks (CNN).

[0175] In some embodiments, the preset model may include a first preset network, a second preset network, and a third preset network. The first preset network, the second preset network, and the third preset network share the same network parameters. Step (5) "training the preset model with the training data until the preset model converges to obtain a feature extraction network" may include the following steps:

[0176] A. Extract the feature vectors of the anchor samples using a first preset network;

[0177] B. Extract the feature vectors of the positive samples using a second preset network;

[0178] C. Extract the feature vectors of the negative samples using a third preset network;

[0179] D. Update the network parameters of the first preset network, the second preset network, and the third preset network according to the feature vectors of the anchor samples, the feature vectors of the positive samples, and the feature vectors of the negative samples until the preset model converges, and obtain three feature extraction networks with the same network parameters.

[0180] For example, referring to Figure 1f , the first preset network, the second preset network, and the third preset network can all be CNNs, and the network parameters are shared among each CNN, that is, the network parameters are the same. The feature vectors of the anchor samples, the feature vectors of the positive samples, and the feature vectors of the negative samples can calculate the loss value (Loss), and update the preset model based on the loss value until the preset model converges, and obtain three feature extraction networks with the same network parameters.

[0181] Among them, the loss value is a parameter used to evaluate the difference between the predicted value and the true value of the neural network model.

[0182] Among them, the network parameters can include weight parameters and bias parameters. The first preset network, the second preset network, and the third preset network sharing the same network parameters means that: if the network parameters of any preset network in the preset model change to new values, the network parameters of other preset networks also change to these new values.

[0183] Among them, the feature vector refers to an image vector represented in the form of a vector.

[0184] In some embodiments, in order to further improve the accuracy of image discrimination, step D "update the network parameters of the first preset network, the second preset network, and the third preset network according to the feature vectors of the anchor samples, the feature vectors of the positive samples, and the feature vectors of the negative samples until the preset model converges, and obtain three feature extraction networks with the same network parameters" may include the following steps:

[0185] Calculate the vector distance between the feature vector of the anchor sample and the feature vector of the positive sample to obtain a first distance;

[0186] Calculate the vector distance between the feature vector of the anchor sample and the feature vector of the negative sample to obtain a second distance;

[0187] Calculate the loss value, and the loss value is the sum of the distance difference and a preset threshold, and the distance difference is the difference between the first distance and the second distance;

[0188] Update the network parameters of the first preset network, the second preset network, and the third preset network using the loss value until the loss value is not greater than 0, and obtain three feature extraction networks with the same network parameters.

[0189] For example, the calculation formula of the loss value Loss is as follows:

[0190] Loss = max(d(a, p) - d(a, n) + margin, 0)

[0191] Where a is the anchor sample, p is the positive sample, n is the negative sample, d(a, p) is the first distance, d(a, n) is the second distance, and margin is the preset threshold.

[0192] For example, referring to Figure 1g , in some embodiments, the network parameters are adjusted by the loss value, so that the first distance between the feature vectors of the anchor sample and the positive sample becomes smaller and smaller, and the second distance between the feature vectors of the anchor sample and the negative sample becomes larger and larger, so that the difference between the same classes becomes smaller and the difference between different classes becomes larger.

[0193] The training method of the feature extraction network is introduced above. Next, how to obtain the visual similarity between the design drawing and the rendering drawing based on the feature extraction network trained above will be introduced:

[0194] In some embodiments, step 103 may include the following steps:

[0195] (1) Construct a similarity calculation model, which may include two feature extraction networks with the same network parameters;

[0196] (2) Use one of the feature extraction networks in the similarity calculation model to calculate the feature vector of the design drawing;

[0197] (3) Use the other feature extraction network in the similarity calculation model to calculate the feature vector of the rendering drawing;

[0198] (4) Take the vector distance between the feature vector of the design drawing and the feature vector of the rendering drawing as the visual similarity between the design drawing and the rendering drawing.

[0199] For example, referring to Figure 1h , input the design drawing of the game firearm into the feature extraction network to obtain the feature vector of the design drawing, and input the rendering drawing of the game firearm into another feature extraction network to obtain the feature vector of the rendering drawing, and calculate the vector distance therebetween as the visual similarity.

[0200] The network structure of the feature extraction network will be introduced in detail below, so it will not be elaborated here.

[0201] It should be noted that, in addition to the preset first angle and the preset second angle, in some embodiments, the game model can also be drawn in the game based on multiple angles such as the preset third angle, the preset fourth angle, and the preset fifth angle, etc., to obtain multiple images such as the third drawing, the fourth drawing, and the fifth drawing of the game model. The method of using the third drawing, the fourth drawing, and the fifth drawing as training data to train the preset model is similar to that of the first and second drawings, so it will not be elaborated here.

[0202] 104. When the visual similarity is within the preset range, the discrimination result is determined as that the target model is drawn normally in the game.

[0203] For example, the preset range can be (0, 0.3). If the visual similarity calculated in step 103 is between 0 and 0.3, the design drawing and the drawing are similar, so the discrimination result can be determined as that the target model is drawn normally in the game.

[0204] 105. When the visual similarity is not within the preset range, the discrimination result is determined as that the target model is drawn abnormally in the game.

[0205] For example, the preset range can be (0, 0.3). If the visual similarity calculated in step 103 is not between 0 and 0.3, the design drawing and the drawing are not similar, so the discrimination result can be determined as that the target model is drawn abnormally in the game.

[0206] 106. Display the discrimination result of the target model to the user.

[0207] In some embodiments, the discrimination result of the target model can be sent to the client, and the client displays the discrimination result of the target model to the user.

[0208] As can be seen from the above, the embodiments of the present application can obtain the appearance design drawing of the target item, where the target item is a game item; based on the preset first angle or the preset second angle, draw the target item in the game to obtain the drawing of the target item; obtain the visual similarity between the appearance design drawing and the drawing; when the visual similarity is within the preset range, determine the discrimination result as that the target model is drawn normally in the game; when the visual similarity is not within the preset range, determine the discrimination result as that the target model is drawn abnormally in the game; display the discrimination result to the user. This solution can automatically discriminate the difference between the game model and the appearance design drawing at different angles in the game, with fast discrimination speed and high accuracy. Therefore, this solution can improve the accuracy of image discrimination.

[0209] The method described in the above embodiments will be further described in detail below.

[0210] The image discrimination solution provided by the embodiments of this application can be applied to the testing of various virtual scenarios. For example, taking the testing of the appearance design drawing of a game firearm and its rendering effect in the game as an example, after adopting the image discrimination solution of the embodiments of this application, there is no need for manual repeated testing, which saves manpower and speeds up the discrimination speed. Therefore, adopting the solution provided by the embodiments of this application can more efficiently and accurately implement the game firearm test.

[0211] In this embodiment, taking the server providing the service of testing the appearance design drawing of a game firearm and its rendering effect in the game for the client as an example, the method of the embodiments of this application will be described in detail.

[0212] As Figure 2a shown, the discrimination method for abnormal rendering of this model is applicable to the server, and the specific process of the discrimination method for abnormal rendering of this model is as follows:

[0213] 201. Obtain a list of firearm names from the client, and the list of firearm names may include the names of game firearms.

[0214] In some embodiments, an accessory loading instruction may also be obtained from the client.

[0215] In some embodiments, the list of firearm names may be a planning configuration table, which is a document in a specified format and may contain firearm information such as names, coatings, and assembled accessories.

[0216] 202. Based on the name of the target firearm, search for the appearance design drawing of the target firearm in the design drawing database. [[ID=2,3]]

[0217] In some embodiments, the target accessory may also be searched for in the game library according to the accessory loading instruction, and the target accessory may be loaded on the target firearm to obtain the loaded target firearm.

[0218] Among them, the target firearm is a game firearm, and the game firearm refers to the firearm model in the game.

[0219] Among them, the target accessory is a game accessory, and the game accessory refers to the firearm accessory model in the game.

[0220] In some embodiments, the design drawing database may be a database server, or a database carried in the client, or a database carried in the server.

[0221] In some embodiments, the game library may be a server running a game program, or a game program carried in the client, or a game program carried in the server.

[0222] 203. Based on a preset first angle or a preset second angle, perform rendering processing on a target firearm in the game to obtain a rendering of the target firearm.

[0223] For example, in some embodiments, based on a preset first angle or a preset second angle, perform rendering processing on the loaded target firearm in the game to obtain a rendering of the target firearm.

[0224] For instance, in some embodiments, the preset first angle may be 0°, and the preset second angle may be 180°. Refer to Figure 2b , Figure 2b in which (i) is the design drawing of rifle 001, Figure 2b in which (ii) is the firearm preview interface of the game. The left side of rifle 001 is defaultly shown in this firearm preview interface, and the user can perform custom settings on the firearm in this interface. For example, add or remove accessories for the firearm, change the paint job of the firearm, view information such as the attributes and price of the firearm.

[0225] Refer to Figure 2b in which (ii). In this firearm preview interface, rifle 001 is equipped with an optical sight (a type of sight accessory), a vertical grip (a type of grip accessory), and a flash hider muzzle (a type of muzzle). By rotating the loaded rifle 001 by 0° in this firearm preview interface, a rendering of the left side of rifle 001 as shown in Figure 2b (iii) can be obtained; or, the loaded rifle 001 can be rotated by 180° in this firearm preview interface, and a rendering of the right side of rifle 001 as shown in Figure 2b (iv) can be obtained.

[0226] Among them, the paint job refers to covering a protective layer or a decorative layer on the surface of the firearm. The concept of firearm paint job in the game is similar to that of game skins, and it is a decorative item used to modify the appearance of the model. For example, firearm paint jobs can include desert camouflage paint job, carbon black paint job, mirror paint job, ebony paint job, rainforest camouflage paint job, and so on.

[0227] In some embodiments, in addition to the artistic representation of the structure and shape of the firearm in the design drawing, the paint job of the firearm can also be shown. For example, Figure 2c in which (i) is rifle 001 with a carbon black paint job, Figure 2c in which (ii) is rifle 001 with an ebony paint job.

[0228] 204. Calculate the visual similarity between the design drawing and the rendering.

[0229] In some embodiments, training data can be prepared in advance, and the preset model can be trained using the training data until the preset model converges to obtain a feature extraction network.

[0230] The steps for preparing the training data will be introduced below:

[0231] (1) Obtain the design drawing of the firearm to be processed;

[0232] (2) Perform drawing processing on the firearm to be processed in the game based on 0°, and obtain the left-side drawing of the firearm to be processed;

[0233] (3) Perform drawing processing on the firearm to be processed in the game based on 180°, and obtain the right-side drawing of the firearm to be processed;

[0234] (4) Add anchor point labels to the design drawing of the firearm to be processed to obtain anchor point samples; add positive labels to the left-side drawing or right-side drawing of the firearm to be processed to obtain positive samples; add negative labels to the left-side drawing or right-side drawing of other firearms to obtain negative samples;

[0235] (5) Use the training data to train the preset model until the preset model converges to obtain the feature extraction network.

[0236] The left-side drawing and right-side drawing obtained by the above method can completely display the features such as the shape and color of the firearm, so as to facilitate the maximum comparison of differences.

[0237] Among them, the firearm to be processed refers to the firearm model of the game firearm, and other firearms refer to the firearm models of other game firearms except the firearm to be processed.

[0238] Among them, the training data can include a triple of anchor point samples, positive samples, and negative samples (anchor point samples, positive samples, negative samples).

[0239] Since sights, muzzles, grips, rifling, ornaments, etc. have little impact on the visual performance of firearms, while stocks, coatings, etc. can greatly affect the visual performance such as the shape and color of firearms. Therefore, referring to Table 2, in some embodiments, if any one of the model, coating, stock, etc. of the firearm in the drawing is different from that of the firearm in the design drawing, then the drawing is not similar to the design drawing, and it can be determined that the discrimination result is that the target model is abnormally drawn in the game. Therefore, the drawing is labeled as a negative sample.

[0240] Otherwise, the drawing is similar to the design drawing, and it can be determined that the target model is normally drawn in the game. Therefore, the drawing is labeled as a positive sample.

[0241] Table 2

[0242] Whether it Affects the Judgment of Similarity Model Yes Sight Accessories No Muzzle Accessories No Grip Accessories No Rifling Accessories No Stock Accessories Yes Coating Yes Ornament No

[0243] Among them, the learning rate during the training process can be set to 0.0001, the batchsize can be set to 32, and the epoch can be set to 32, that is, the training stops after 32 updates. Among them, an epoch refers to the process where a complete dataset passes through the neural network once and returns once, and this process is called one epoch.

[0244] In the embodiments of the present application, as Figure 2d shown, during the training process, the curve of the loss value Loss gradually decreases and tends to be dynamically stable as the epoch increases for the training data (train) and the test data (validation), and there is no overfitting or underfitting.

[0245] Then, the trained feature extraction network can be used to determine the visual similarity between the design drawing and the rendering drawing. For example, in some embodiments, the step "calculating the feature vector of the design drawing using a feature extraction network in the similarity calculation model" may include the following steps:

[0246] (1) Construct a similarity calculation model, which may include two feature extraction networks with the same network parameters;

[0247] (2) Calculate the feature vector of the design drawing using a feature extraction network in the similarity calculation model;

[0248] (3) Calculate the feature vector of the rendering drawing using the other feature extraction network in the similarity calculation model;

[0249] (4) Use the vector distance between the feature vector of the design drawing and the feature vector of the rendering drawing as the visual similarity between the design drawing and the rendering drawing.

[0250] In some embodiments, the feature extraction network can introduce the network structure of Residual Networks (ResNets), which may include convolutional layers, and the convolutional layers may include multiple residual blocks. Therefore, the step (3) "calculating the feature vector of the design drawing using a feature extraction network in the similarity calculation model" may include the following steps:

[0251] Determine the output result of the previous residual block and the current residual block located after the previous residual block;

[0252] Perform convolutional processing on the output result of the previous residual block in the current residual block to obtain a convolutional result;

[0253] Add the convolutional result and the output result of the previous residual block to obtain the output result of the current residual block;

[0254] Calculate the feature vector of the design drawing based on the convolutional matrix, where the convolutional matrix is the output result of the last residual block.

[0255] Reference Figure 2e , in some embodiments, Figure 2e is the structure of each residual block in the residual network, and the residual block can be expressed as:

[0256] X i+1 = X i + F(X i , W i )

[0257] where X i is the output result of the previous residual block, X i+1 is the output result of the current residual block, F(X i , W i ) is the convolution result obtained by performing convolution processing on the output result of the previous residual block in the current residual block, and W i is the weight in the weight layer, and the weight layer can include two layers, and a rectified linear unit (ReLU) can be included between the layers. The ReLU function formula is as follows:

[0258]

[0259] where f(x) is the output, x is the input, and λ is a constant.

[0260] In some embodiments, in order to reduce the calculation time and the number of parameters while ensuring the accuracy, the feature extraction network can introduce a depthwise separable convolution structure. The depthwise separable convolution is a kind of factorized convolutions. As Figure 2f shown, the depthwise separable convolution divides the traditional convolution operation ( Figure 2f (i)) into two steps, namely the depthwise convolution operation ( Figure 2f [[ID= 46]](ii)) and the pointwise convolution operation ( Figure 2f (iii)).

[0261] For the standard convolution, its convolution kernel is used on all input channels, while there is a mapping relationship between the depthwise convolution and the input channels, and the pointwise convolution is an ordinary 1x1 convolution kernel.

[0262] For example, as Figure 2fAs shown, M represents the number of channels of the input features, and N represents the number of channels of the output features (which is also the number of convolutional kernels of this layer). Therefore, if it is assumed that the size of the convolutional kernel is DK*DK*M*N and the output is DF*DF*N, then the computational complexity of the standard convolution is DK*DK*M*N*DF*DF.

[0263] Therefore, in some embodiments, the feature extraction network may include a convolutional layer, and the convolutional layer includes depthwise convolution. Therefore, the step of "calculating the feature vector of the design drawing by using a feature extraction network in the similarity calculation model" may include the following steps:

[0264] (1) Obtain a preset mapping relationship, which is used to describe the depthwise convolution corresponding to each channel. The design drawing has multiple channels.

[0265] (2) Use the depthwise convolution corresponding to the channel to perform convolution processing on the design drawing of the channel to obtain the depthwise convolution result of the channel.

[0266] (3) Integrate the depthwise convolution results of all channels into a convolution matrix.

[0267] (4) Calculate the feature vector of the design drawing based on the convolution matrix.

[0268] There are various ways to perform the step (3) of "integrating the depthwise convolution results of all channels into a convolution matrix". For example, in some embodiments, in order to prevent model overfitting and prompt the model to learn the essential information of the feature vector, the step (3) of "integrating the depthwise convolution results of all channels into a convolution matrix" may include the following steps:

[0269] Perform sharding processing on the depthwise convolution result of the channel to obtain multiple result segments of the channel.

[0270] Perform shuffling processing on the multiple result segments of all channels to obtain the shuffled result segments.

[0271] Integrate the shuffled result segments into a convolution matrix.

[0272] Calculate the feature vector of the design drawing based on the convolution matrix.

[0273] For example, perform sharding on the depthwise convolution result of each channel to obtain multiple result segments of a preset size, and then perform random sorting on the result segments to shuffle them into shuffled result segments. Then, splice the shuffled result segments into a convolution matrix to achieve integration, improving the robustness of the network model.

[0274] Since the feature extraction network in step (3) "calculating the feature vector of the design drawing using a feature extraction network in the similarity calculation model" is the same as the feature extraction network in step (4) "using the vector distance between the feature vector of the design drawing and the feature vector of the drawing as the visual similarity between the design drawing and the drawing", and the implementation steps are similar, the implementation of step (4) "using the vector distance between the feature vector of the design drawing and the feature vector of the drawing as the visual similarity between the design drawing and the drawing" will not be elaborated here.

[0275] Reference Figure 2g , in some embodiments, the convolutional matrix may include multiple matrix elements, and the feature extraction network further includes a weight ignoring layer, a hidden layer, a normalization layer, and an embedding layer. The step "calculating the feature vector of the design drawing based on the convolutional matrix" may include the following steps:

[0276] Among them, the weight ignoring layer may include a drop-out layer, which can randomly ignore the weight update of some hidden layer nodes in the neural network during model training, thereby improving the training efficiency and preventing overfitting.

[0277] The weight ignoring layer randomly selects matrix elements from the convolutional matrix as elements to be set to zero, and performs zeroing processing on the elements to be set to zero in the convolutional matrix to obtain a processed convolutional matrix;

[0278] The hidden layer performs dimensionality conversion processing on the processed convolutional matrix to obtain a hidden layer result;

[0279] The normalization layer performs normalization processing on the hidden layer result to obtain a normalized result;

[0280] The embedding layer performs vector embedding processing on the normalized result to obtain the feature vector of the design drawing.

[0281] Finally, the visual similarity is calculated based on the feature vector of the design drawing and the feature vector of the drawing.

[0282] 205. Obtain a discrimination result according to the visual similarity.

[0283] For example, when the visual similarity is within a preset range, the discrimination result is determined as the target model being normally drawn in the game; when the visual similarity is not within the preset range, the discrimination result is determined as the target model being abnormally drawn in the game.

[0284] 206. Send the design drawing and drawing of the target firearm and the discrimination result to the client, so that the client can display the discrimination result to the user.

[0285] For example, reference Figure 2h , the client can display the discrimination result on the page.

[0286] The Receiver Operating Characteristic (ROC) curve is a graphical analysis tool used to select the best signal detection model, discard sub-optimal models, or set the optimal threshold within the same model.

[0287] Among them, the horizontal axis of the ROC curve is the false positive rate (FPR), which represents the proportion of actual negative samples among the positive samples predicted by the classifier in all negative samples; the vertical axis of the ROC curve is the true positive rate (TPR) sensitivity, which represents the proportion of actual positive samples among the positive samples predicted by the classifier in all positive samples.

[0288] When TPR = 1 and FPR = 0 (i.e., the point (0, 1) in the figure), the closer the ROC curve is to the point (0, 1) and the more deviated from the 45-degree diagonal line, the better the network model.

[0289] The area under the ROC curve is AUC (Area under the Curve of ROC, AUC). The larger the AUC value, the higher the accuracy of image discrimination.

[0290] In the embodiments of the present application, as Figure 2i shown, the AUC of the image discrimination model in the embodiments of the present application reaches 0.99, and the accuracy of image discrimination is high.

[0291] As can be seen from the above, the embodiments of the present application can obtain a list of firearm names from the client, and the list of firearm names can include the firearm names of game firearms; based on the firearm name of the target firearm, search for the external design drawing of the target firearm in the design drawing database; obtain an accessory loading instruction from the client; search for the target accessory in the game library according to the accessory loading instruction; load the target accessory onto the target firearm to obtain the loaded target firearm; perform a drawing process on the loaded target firearm in the game based on a preset first angle or a preset second angle to obtain a drawing of the target firearm; calculate the visual similarity between the external design drawing and the drawing; obtain a discrimination result according to the visual similarity; when the visual similarity is within a preset range, determine that the discrimination result is that the target model is drawn normally in the game; when the visual similarity is not within the preset range, determine that the discrimination result is that the target model is drawn abnormally in the game; send the external design drawing and the drawing of the target firearm and the discrimination result to the client so that the client can display the discrimination result to the user.

[0292] Therefore, the embodiments of the present application can centrally provide image discrimination services for multiple users by the server. The discrimination speed and accuracy are high, and during the game production process, it can test and give early warnings for the situation where the drawn image has a large difference from the originally designed appearance design drawing, effectively solving the problem that the game model in the game screen does not match its original appearance design drawing only after the game is released.

[0293] Therefore, the embodiments of the present application can efficiently and accurately discriminate the similarity between the appearance design drawing and the drawn image, so as to discover and solve problems in a timely manner.

[0294] To better implement the above method, the embodiments of the present application further provide a discrimination device for abnormal model drawing. The discrimination device for abnormal model drawing can be specifically integrated in an electronic device, and the electronic device can be a device such as a terminal or a server. Among them, the terminal can be a device such as a mobile phone, a tablet computer, a smart Bluetooth device, a notebook computer, or a personal computer; the server can be a single server or a server cluster composed of multiple servers.

[0295] For example, in this embodiment, taking the discrimination device for abnormal model drawing specifically integrated in the server as an example, the method of the embodiments of the present application will be described in detail.

[0296] For example, as Figure 3 shown, the discrimination device for abnormal model drawing may include an acquisition unit 301, a drawing unit 302, a similarity unit 303, a normal unit 304, an abnormal unit 305, and a display unit 306, as follows:

[0297] (1) Acquisition unit 301.

[0298] The acquisition unit 301 can be used to acquire the appearance design drawing of the target model, and the target model is a game model.

[0299] In some embodiments, the acquisition unit 301 can be used to:

[0300] Acquire a list of model names, and the list of model names may include the model names of game models;

[0301] Based on the model name of the target model, search for the appearance design drawing of the target model in the design drawing database;

[0302] Showing the discrimination result of the target model to the user may be sending the appearance design drawing, the drawn image, and the discrimination result of the target model to the client, so that the client can show the discrimination result to the user.

[0303] (2) Drawing unit 302.

[0304] The drawing unit 302 can be used to perform drawing processing on the target model in the game based on a preset first angle or a preset second angle to obtain a drawing of the target model.

[0305] In some embodiments, the drawing unit 302 can be used for:

[0306] Obtain an accessory loading instruction;

[0307] Search for the target accessory in the game model library according to the accessory loading instruction;

[0308] Load the target accessory onto the target model to obtain the loaded target model;

[0309] Based on a preset first angle or a preset second angle, perform drawing processing on the loaded target model in the game to obtain a drawing of the target model.

[0310] (3) Similarity unit 303.

[0311] The similarity unit 303 can be used to obtain the visual similarity between the design drawing and the drawing.

[0312] In some embodiments, the similarity unit 303 can include a design drawing subunit, a first drawing subunit, a second drawing subunit, a data subunit, a training subunit, and a similarity subunit, as follows:

[0313] (1) Design drawing subunit.

[0314] The design drawing subunit can be used to obtain the design drawing of the game model.

[0315] (2) First drawing subunit.

[0316] The first drawing subunit can be used to perform drawing processing on the game model in the game based on a preset first angle to obtain the first drawing of the game model.

[0317] (3) Second drawing subunit.

[0318] The second drawing subunit can be used to perform drawing processing on the game model in the game based on a preset second angle to obtain the second drawing of the game model.

[0319] (4) Data subunit.

[0320] The data subunit can be used to obtain training data based on the design drawing, the first drawing, and the second drawing of the game model.

[0321] In some embodiments, the game model may include a model to be processed and other models. The other models are game models other than the model to be processed. The data subunit may include an anchor point sub-module, a positive sample sub-module, a negative sample sub-module, and a data sub-module, as follows:

[0322] A. Anchor point sub-module.

[0323] The anchor point sub-module can be used to add anchor point labels to the appearance design drawing of the model to be processed to obtain anchor point samples.

[0324] B. Positive sample sub-module.

[0325] The positive sample sub-module can be used to add positive labels to the first drawing or the second drawing of the model to be processed to obtain positive samples.

[0326] C. Negative sample sub-module.

[0327] The negative sample sub-module can be used to add negative labels to the first drawing or the second drawing of other models to obtain negative samples.

[0328] D. Data sub-module.

[0329] The data sub-module can be used to obtain training data, and the training data may include anchor point samples, positive samples, and negative samples.

[0330] (5) Training subunit.

[0331] The training subunit can be used to train a preset model with the training data until the preset model converges to obtain a feature extraction network.

[0332] In some embodiments, the preset model may include a first preset network, a second preset network, and a third preset network. The first preset network, the second preset network, and the third preset network share the same network parameters. The training subunit may include an anchor point vector sub-module, a positive sample vector sub-module, a negative sample vector sub-module, and an update sub-module, as follows:

[0333] E. The anchor point vector sub-module can be used to extract the feature vector of the anchor point sample by using the first preset network;

[0334] F. The positive sample vector sub-module can be used to extract the feature vector of the positive sample by using the second preset network;

[0335] G. The negative sample vector sub-module can be used to extract the feature vector of the negative sample by using the third preset network;

[0336] The H. update sub-module can be used to update the network parameters of the first preset network, the second preset network, and the third preset network according to the feature vectors of the anchor samples, the feature vectors of the positive samples, and the feature vectors of the negative samples, until the preset model converges, and obtain three feature extraction networks with the same network parameters.

[0337] In some embodiments, the update sub-module can be used to:

[0338] Calculate the vector distance between the feature vector of the anchor sample and the feature vector of the positive sample to obtain a first distance;

[0339] Calculate the vector distance between the feature vector of the anchor sample and the feature vector of the negative sample to obtain a second distance;

[0340] Calculate the loss value, which is the sum of the distance difference and the preset threshold, and the distance difference is the difference between the first distance and the second distance;

[0341] Use the loss value to update the network parameters of the first preset network, the second preset network, and the third preset network until the loss value is not greater than 0, and obtain three feature extraction networks with the same network parameters.

[0342] (6) Similarity sub-unit.

[0343] The similarity sub-unit can be used to determine the visual similarity between the design drawing and the rendering drawing using the feature extraction network.

[0344] For example, in some embodiments, the similarity unit 303 may include a construction sub-unit, a design drawing vector sub-unit, a rendering drawing vector sub-unit, and a distance sub-unit, as follows:

[0345] (1) Construction sub-unit.

[0346] The construction sub-unit can be used to construct a similarity calculation model, and the similarity calculation model can include two feature extraction networks with the same network parameters.

[0347] (2) Design drawing vector sub-unit.

[0348] The design drawing vector sub-unit can be used to calculate the feature vector of the design drawing using one of the feature extraction networks in the similarity calculation model.

[0349] In some embodiments, the feature extraction network may include a convolutional layer, the convolutional layer may include multiple residual blocks, and the design drawing vector sub-unit can be used to:

[0350] Determine the output result of the previous residual block and the current residual block located after the previous residual block;

[0351] Perform convolution processing on the output result of the previous residual block in the current residual block to obtain a convolution result;

[0352] Add the convolution result and the output result of the previous residual block to obtain the output result of the current residual block;

[0353] Calculate the feature vector of the design drawing based on the convolution matrix, where the convolution matrix is the output result of the last residual block.

[0354] In some embodiments, the feature extraction network may include a convolution layer, and the convolution layer may include depth convolution. The design drawing vector sub-unit may include a mapping sub-module, a depth convolution sub-module, an integration sub-module, and a design drawing vector sub-module, as follows:

[0355] A. Mapping sub-module.

[0356] The mapping sub-module can be used to obtain a preset mapping relationship, and the mapping relationship can be used to describe the depth convolution corresponding to each channel. The design drawing has multiple channels.

[0357] B. Depth convolution sub-module.

[0358] The depth convolution sub-module can be used to perform convolution processing on the design drawing of the channel by using the depth convolution corresponding to the channel to obtain the depth convolution result of the channel.

[0359] C. Integration sub-module.

[0360] The integration sub-module can be used to integrate the depth convolution results of all channels into a convolution matrix.

[0361] In some embodiments, the integration sub-module can be used to:

[0362] Perform slicing processing on the depth convolution result of the channel to obtain multiple result segments of the channel;

[0363] Perform shuffling processing on the multiple result segments of all channels to obtain the shuffled result segments;

[0364] Integrate the shuffled result segments into a convolution matrix;

[0365] Calculate the feature vector of the design drawing based on the convolution matrix.

[0366] D. Design drawing vector sub-module.

[0367] The design drawing vector sub-module can be used to calculate the feature vector of the design drawing based on the convolution matrix.

[0368] (3) Draw the drawing vector sub-unit.

[0369] The drawing graph vector sub-unit can be used to calculate the feature vector of the drawing graph by using another feature extraction network in the similarity calculation model.

[0370] (4) Distance sub-unit.

[0371] The distance sub-unit can be used to take the vector distance between the feature vector of the design drawing and the feature vector of the drawing graph as the visual similarity between the design drawing and the drawing graph.

[0372] In some embodiments, the convolutional matrix may include a plurality of matrix elements, and the feature extraction network may further include a weight ignoring layer, a hidden layer, a normalization layer, and an embedding layer. When the design drawing vector sub-unit is used to calculate the feature vector of the design drawing based on the convolutional matrix, it can specifically be used for:

[0373] In the weight ignoring layer, randomly select matrix elements from the convolutional matrix as the elements to be set to zero, and perform zeroing processing on the elements to be set to zero in the convolutional matrix to obtain a processed convolutional matrix;

[0374] In the hidden layer, perform dimensionality conversion processing on the processed convolutional matrix to obtain a hidden layer result;

[0375] In the normalization layer, perform normalization processing on the hidden layer result to obtain a normalized result;

[0376] In the embedding layer, perform vector embedding processing on the normalized result to obtain the feature vector of the design drawing.

[0377] (IV) Normal unit 304.

[0378] The normal unit 304 can be used to determine that the discrimination result is that the target model is drawn normally in the game when the visual similarity is within a preset range.

[0379] (V) Abnormal unit 305.

[0380] The abnormal unit 305 can be used to determine that the discrimination result is that the target model is drawn abnormally in the game when the visual similarity is not within the preset range.

[0381] (VI) Display unit 306.

[0382] The display unit 306 can be used to display the discrimination result of the target model to the user.

[0383] In specific implementation, the above-mentioned each unit can be implemented as an independent entity, or can be combined arbitrarily and implemented as the same or several entities. For the specific implementation of the above-mentioned each unit, reference can be made to the foregoing method embodiments, which will not be elaborated herein.

[0384] As described above, the discriminant device for abnormal rendering of the model in this embodiment obtains the design drawing of the target model through an acquisition unit, where the target model is a game model; a rendering unit performs rendering processing on the target model in the game based on a preset first angle or a preset second angle to obtain a rendering of the target model; a similarity unit obtains the visual similarity between the design drawing and the rendering; a normal unit determines that the discriminant result is that the design drawing and the rendering are similar when the visual similarity is within a preset range; an abnormal unit determines that the discriminant result is that the target model is abnormally rendered in the game when the visual similarity is not within the preset range; and a display unit displays the discriminant result of the target model to the user.

[0385] Thus, the embodiment of the present application can efficiently discriminate the similarity between the design drawing and the rendering, improving the accuracy of image discrimination.

[0386] The embodiment of the present application also provides an electronic device, which can be a device such as a terminal, a server, etc. Among them, the terminal can be a mobile phone, a tablet computer, a smart Bluetooth device, a laptop computer, a personal computer, etc.; the server can be a single server or a server cluster composed of multiple servers, etc.

[0387] In some embodiments, the discriminant device for abnormal rendering of the model can also be integrated in multiple electronic devices. For example, the discriminant device for abnormal rendering of the model can be integrated in multiple servers, and the discriminant method for abnormal rendering of the model of the present application is implemented by multiple servers.

[0388] In this embodiment, the electronic device in this embodiment is taken as an example of a server for detailed description. For example, as Figure 4 shown, it shows a schematic structural diagram of the server involved in the embodiment of the present application. Specifically:

[0389] The server may include a processor 401 with one or more processing cores, a memory 402 with one or more computer-readable storage media, a power supply 403, an input module 404, and a communication module 405 and other components. Those skilled in the art can understand that Figure 4 the server structure shown in

[0390] The processor 401 is the control center of the server, connecting various parts of the entire server through various interfaces and lines. By running or executing software programs and / or modules stored in the memory 402, and by invoking the data stored in the memory 402, it performs various functions of the server and processes data, thereby monitoring the server as a whole. In some embodiments, the processor 401 may include one or more processing cores; in some embodiments, the processor 401 may integrate an application processor and a modem processor. Among them, the application processor mainly processes the operating system, user interface, application programs, etc., and the modem processor mainly processes wireless communications. It can be understood that the above-mentioned modem processor may not be integrated into the processor 401 either.

[0391] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402. The memory 402 mainly includes a program storage area and a data storage area. Among them, the program storage area can store the operating system, application programs required for at least one function (such as the sound playback function, image playback function, etc.); the data storage area can store data created according to the use of the server. In addition, the memory 402 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage devices. Correspondingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.

[0392] The server further includes a power supply 403 that powers each component. In some embodiments, the power supply 403 can be logically connected to the processor 401 through a power management system, so as to implement functions such as management of charging, discharging, and power consumption management through the power management system. The power supply 403 may also include any components such as one or more DC or AC power supplies, a recharge system, a power failure detection circuit, a power converter or inverter, and a power status indicator.

[0393] The server may further include an input module 404, which can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function controls.

[0394] The server may further include a communication module 405. In some embodiments, the communication module 405 may include a wireless module. The server can perform short-range wireless transmission through the wireless module of the communication module 405, thereby providing users with wireless broadband Internet access. For example, the communication module 405 can be used to help users send and receive emails, browse web pages, and access streaming media, etc.

[0395] Although not shown, the server may also include a display unit and so on, which will not be elaborated here. Specifically, in this embodiment, the processor 401 in the server will load the executable files corresponding to the processes of one or more application programs into the memory 402 according to the following instructions, and the processor 401 will run the application programs stored in the memory 402 to implement various functions as follows:

[0396] Obtain the design drawing of the target model, where the target model is a game model;

[0397] Based on a preset first angle or a preset second angle, perform rendering processing on the target model in the game to obtain a rendering of the target model;

[0398] Obtain the visual similarity between the design drawing and the rendering;

[0399] When the visual similarity is within the preset range, determine the discrimination result as that the target model is rendered normally in the game;

[0400] When the visual similarity is not within the preset range, determine the discrimination result as that the target model is rendered abnormally in the game;

[0401] Show the discrimination result of the target model to the user.

[0402] For the specific implementation of each of the above operations, reference may be made to the previous embodiments, which will not be elaborated here.

[0403] As can be seen from the above, the present application can efficiently discriminate the similarity between the design drawing and the rendering, and improve the accuracy of image discrimination.

[0404] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions, or by controlling related hardware through instructions. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0405] Therefore, an embodiment of the present application provides a computer-readable storage medium, in which multiple instructions are stored. These instructions can be loaded by a processor to execute the steps in any method for discriminating abnormal rendering of a model provided by the embodiments of the present application. For example, these instructions can execute the following steps:

[0406] Obtain the design drawing of the target model, where the target model is a game model;

[0407] Based on a preset first angle or a preset second angle, perform rendering processing on the target model in the game to obtain a rendering of the target model;

[0408] Obtain the visual similarity between the design drawing and the rendering;

[0409] When the visual similarity is within the preset range, the discrimination result is determined as the target model being normally rendered in the game;

[0410] When the visual similarity is not within the preset range, the discrimination result is determined as the target model being abnormally rendered in the game;

[0411] Show the discrimination result of the target model to the user.

[0412] Among them, the storage medium may include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc.

[0413] According to one aspect of the present application, there is provided a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in various alternative implementations of the image discrimination aspect or the game test aspect provided in the above embodiments.

[0414] Since the instructions stored in the storage medium can execute the steps in any of the discrimination methods for abnormal model rendering provided in the embodiments of the present application, therefore, the beneficial effects that can be achieved by any of the discrimination methods for abnormal model rendering provided in the embodiments of the present application can be realized. For details, please refer to the previous embodiments and will not be elaborated here.

[0415] The above has introduced in detail a discrimination method, device, server and computer-readable storage medium for abnormal model rendering provided by the embodiments of the present application. Specific examples are used in this article to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A discriminant method for abnormal model drawing, characterized in that, Including: Obtain the design drawing of the target model, where the target model is a game virtual model, and the design drawing is an art scheme designed for the game virtual model and presented in the form of a two-dimensional image; Obtain an accessory loading instruction, and search for the target accessory in the game model library according to the accessory loading instruction; Load the target accessory onto the target model to obtain the loaded target model; Based on a preset first angle or a preset second angle, perform three-dimensional drawing processing on the loaded target model in the game to obtain a three-dimensional drawing of the loaded target model; Obtain the visual similarity between the design drawing and the three-dimensional drawing; When the visual similarity is within a preset range, determine the discrimination result as that the target model is drawn normally in the game; When the visual similarity is not within the preset range, determine the discrimination result as that the target model is drawn abnormally in the game; Display the discrimination result of the target model to the user.

2. The discriminant method for abnormal model drawing according to claim 1, wherein The obtaining the visual similarity between the design drawing and the drawing includes: Obtain the design drawing of the game virtual model; Based on a preset first angle, perform drawing processing on the game virtual model in the game to obtain a first drawing of the game virtual model; Based on a preset second angle, perform drawing processing on the game virtual model in the game to obtain a second drawing of the game virtual model; Obtain training data based on the design drawing, the first drawing, and the second drawing of the game virtual model; Use the training data to train a preset model until the preset model converges to obtain a feature extraction network; Use the feature extraction network to determine the visual similarity between the design drawing and the drawing.

3. The discriminant method for abnormal model drawing according to claim 2, wherein The game virtual model includes a model to be processed and other models. The other models are other game virtual models except the model to be processed. The obtaining training data based on the design drawing, the first drawing, and the second drawing of the game virtual model includes: Add anchor point labels to the design drawing of the model to be processed to obtain anchor point samples; Add positive labels to the first drawing or the second drawing of the model to be processed to obtain positive samples; Add negative labels to the first drawing or the second drawing of the other models to obtain negative samples; Obtain training data, where the training data includes anchor point samples, positive samples, and negative samples.

4. The discriminant method for abnormal model drawing according to claim 3, characterized in that The preset model includes a first preset network, a second preset network, and a third preset network. The first preset network, the second preset network, and the third preset network share the same network parameters. The using the training data to train the preset model until the preset model converges to obtain a feature extraction network includes: Use the first preset network to extract the feature vector of the anchor point sample; Use the second preset network to extract the feature vector of the positive sample; Use the third preset network to extract the feature vector of the negative sample; Update the network parameters of the first preset network, the second preset network, and the third preset network according to the feature vectors of the anchor samples, the feature vectors of the positive samples, and the feature vectors of the negative samples until the preset model converges, and obtain three feature extraction networks with the same network parameters.

5. The discrimination method for abnormal model drawing according to claim 4, characterized in that, The step of updating the network parameters of the first preset network, the second preset network, and the third preset network according to the feature vectors of the anchor samples, the feature vectors of the positive samples, and the feature vectors of the negative samples until the preset model converges and obtaining three feature extraction networks with the same network parameters includes: Calculate the vector distance between the feature vector of the anchor sample and the feature vector of the positive sample to obtain a first distance; Calculate the vector distance between the feature vector of the anchor sample and the feature vector of the negative sample to obtain a second distance; Calculate a loss value, where the loss value is the sum of the distance difference and a preset threshold, and the distance difference is the difference between the first distance and the second distance; Use the loss value to update the network parameters of the first preset network, the second preset network, and the third preset network until the loss value is not greater than 0, and obtain three feature extraction networks with the same network parameters.

6. The discrimination method for abnormal drawing of the model according to claim 1, characterized in that, The step of obtaining the visual similarity between the design drawing and the rendering drawing includes: Construct a similarity calculation model, where the similarity calculation model includes two feature extraction networks with the same network parameters; Use one of the feature extraction networks in the similarity calculation model to calculate the feature vector of the design drawing; Use the other feature extraction network in the similarity calculation model to calculate the feature vector of the rendering drawing; Take the vector distance between the feature vector of the design drawing and the feature vector of the rendering drawing as the visual similarity between the design drawing and the rendering drawing.

7. The discrimination method for abnormal model drawing according to claim 6, characterized in that, The feature extraction network includes a convolutional layer, and the convolutional layer includes a plurality of residual blocks. The step of using one of the feature extraction networks in the similarity calculation model to calculate the feature vector of the design drawing includes: Determine the output result of the previous residual block and the current residual block located after the previous residual block; Perform convolution processing on the output result of the previous residual block in the current residual block to obtain a convolution result; Add the convolution result and the output result of the previous residual block to obtain the output result of the current residual block; Calculate the feature vector of the design drawing based on the convolution matrix, where the convolution matrix is the output result of the last residual block.

8. The discriminant method for abnormal drawing of a model according to claim 6, characterized in that, The feature extraction network includes a convolutional layer, and depth convolution is included in the convolutional layer. The step of using one of the feature extraction networks in the similarity calculation model to calculate the feature vector of the design drawing includes: Obtain a preset mapping relationship, where the mapping relationship is used to describe the depth convolution corresponding to each channel, and the design drawing has multiple channels; Use the depth convolution corresponding to the channel to perform convolution processing on the design drawing of the channel to obtain the depth convolution result of the channel; Integrate the depth convolution results of all the channels into a convolution matrix; Calculate the feature vector of the design drawing based on the convolutional matrix.

9. The discriminant method for abnormal drawing of the model according to claim 8, wherein The integration of the depth convolution results of all the channels into a convolutional matrix includes: Perform slicing processing on the depth convolution results of the channels to obtain multiple result segments of the channels; Perform shuffling processing on the multiple result segments of all the channels to obtain the shuffled result segments; Integrate the shuffled result segments into a convolutional matrix; Calculate the feature vector of the design drawing based on the convolutional matrix.

10. The discrimination method for abnormal model drawing according to any one of claims 7 to 9, characterized in that, The convolutional matrix includes multiple matrix elements. The feature extraction network further includes a weight ignoring layer, a hidden layer, a normalization layer, and an embedding layer. The calculation of the feature vector of the design drawing based on the convolutional matrix includes: Randomly select matrix elements from the convolutional matrix as the elements to be set to zero in the weight ignoring layer, and perform zeroing processing on the elements to be set to zero in the convolutional matrix to obtain the processed convolutional matrix; Perform dimension conversion processing on the processed convolutional matrix in the hidden layer to obtain the hidden layer result; Perform normalization processing on the hidden layer result in the normalization layer to obtain the normalized result; Perform vector embedding processing on the normalized result in the embedding layer to obtain the feature vector of the design drawing.

11. The discrimination method for abnormal drawing of the model according to any one of claims 1, characterized in that, The obtaining of the design drawing of the target model includes: Obtain a list of model names, where the list of model names includes the model names of game virtual models; Based on the model name of the target model, search for the design drawing of the target model in the design drawing database; The presenting of the discrimination result of the target model to the user includes: Send the design drawing, the drawing, and the discrimination result of the target model to the client, so that the client presents the discrimination result of the target model to the user.

12. A discriminant device for abnormal model drawing, characterized in that, Includes: An obtaining unit, configured to obtain the design drawing of a target model, where the target model is a game virtual model, and the design drawing is an art scheme designed for the game virtual model and presented in the form of a two-dimensional image; obtain an accessory loading instruction, search for a target accessory in the game model library according to the accessory loading instruction; load the target accessory onto the target model to obtain the loaded target model; A drawing unit, configured to perform drawing processing on the loaded target model in the game based on a preset first angle or a preset second angle to obtain the drawing of the loaded target model; A similarity unit, configured to obtain the visual similarity between the design drawing and the drawing; A normal unit, configured to determine that the discrimination result is that the target model is drawn normally in the game when the visual similarity is within a preset range; An abnormal unit, configured to determine that the discrimination result is that the target model is drawn abnormally in the game when the visual similarity is not within the preset range; A presenting unit, configured to present the discrimination result of the target model to the user.

13. A server, characterized in that, Includes a processor and a memory, where the memory stores multiple instructions; the processor loads the instructions from the memory to execute the steps in the discrimination method for abnormal drawing of the model according to any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores multiple instructions, and the instructions are suitable for being loaded by a processor to execute the steps in the discrimination method for abnormal model drawing according to any one of claims 1 to 11.

15. A computer program product, the computer program product includes computer instructions, and the computer instructions are stored in a computer-readable storage medium; a processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to execute the steps in the discrimination method for abnormal model drawing according to any one of claims 1 to 11.

Images