Content detection method and apparatus for game scenes, and electronic device

Abstract

Provided in the present disclosure are a content detection method and apparatus for game scenes, and an electronic device. The method comprises: acquiring an image to be inspected, wherein a target map model is displayed in said image (S102); inputting said image into a pre-trained map classification model, and processing said image through the map classification model to obtain a processing result (S104); and on the basis of the processing result, determining whether the target map model contains preset specified content (S106). The present disclosure improves the efficiency of reviewing custom game scenes in a game, reduces labor costs, shortens the waiting time of players, and enhances the game experience of the players.

Description

Game scene content detection method, device and electronic equipment

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to Chinese patent application number 202310919043.9, filed on July 25, 2023, entitled “Content detection method, device and electronic device for game scenes”. The entire contents of the Chinese patent application are incorporated herein by reference. Technical Field

[0003] The present disclosure relates to the field of model rendering technology, and in particular to a method, device, and electronic device for detecting content in a game scene. Background Art

[0004] In some games with customization features, players can freely create their own 3D game scenes and interact and communicate with other players. However, due to the game's freedom and openness, some players may exploit the various components provided to construct structures inappropriate for the content, resulting in a degradation of the gaming environment. In related technologies, staff typically review player-submitted custom game scenes to identify those with inappropriate content. However, manual review is time-consuming and inefficient, leading to excessive player wait times and a reduced gaming experience.

[0005] Summary of the Invention

[0006] In view of this, the purpose of the present disclosure is to provide a content detection method, device and electronic device for a game scene, so as to improve the efficiency of reviewing customized game scenes in the game and reduce labor costs.

[0007] In a first aspect, an embodiment of the present disclosure provides a content detection method for a game scene, comprising: obtaining an image to be detected that displays a target game scene; inputting the image to be detected into a pre-trained scene classification model, processing the image to be detected through the scene classification model to obtain a processing result; and determining whether the target game scene contains preset specified content based on the processing result.

[0008] In the second aspect, an embodiment of the present disclosure provides a content detection device for a game scene, comprising: an image acquisition module for detection, configured to acquire an image for detection showing a target game scene; a classification module, configured to input the image for detection into a pre-trained scene classification model, and process the image for detection through the scene classification model to obtain a processing result; and a content determination module, configured to determine whether the target game scene contains preset specified content based on the processing result.

[0009] In a third aspect, an embodiment of the present disclosure provides an electronic device including a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor executes the machine-executable instructions to implement the above-mentioned content detection method for the game scene.

[0010] In a fourth aspect, an embodiment of the present disclosure provides a machine-readable storage medium, which stores machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions prompt the processor to implement the above-mentioned content detection method for the game scene.

[0011] The embodiments of the present disclosure bring the following beneficial effects:

[0012] The aforementioned method, device, and electronic device for detecting game scene content capture an image showing a target game scene; input the image into a pre-trained scene classification model, which processes the image to obtain a processing result; and based on the processing result, determines whether the target game scene contains pre-set, specified content. This approach improves the efficiency of reviewing custom game scenes in games, reduces labor costs, shortens player waiting time, and enhances the player's gaming experience.

[0013] Other features and advantages of the present disclosure will be described in the following description, and in part will become apparent from the description, or understood by practicing the present disclosure. The objectives and other advantages of the present disclosure are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0014] In order to make the above-mentioned objectives, features and advantages of the present disclosure more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the specific embodiments of the present disclosure or the technical solutions in the related technologies, the following briefly introduces the drawings required for use in the specific embodiments or related technical descriptions. Obviously, the drawings described below are some embodiments of the present disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0016] FIG1 is a flow chart of a method for detecting content of a game scene provided by one of the embodiments of the present disclosure;

[0017] FIG2 is a schematic diagram of the structure of a scene classification model provided by one of the embodiments of the present disclosure;

[0018] FIG3 is a schematic diagram of a content detection process provided by one of the embodiments of the present disclosure;

[0019] FIG4 is a schematic diagram of a model classification result provided by one of the embodiments of the present disclosure;

[0020] FIG5 is a schematic structural diagram of a content detection device for a game scene provided by one of the embodiments of the present disclosure;

[0021] FIG6 is a schematic structural diagram of an electronic device provided in accordance with one of the embodiments of the present disclosure. DETAILED DESCRIPTION

[0022] To make the purpose, technical solutions, and advantages of the embodiments of the present disclosure more clear, the technical solutions of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present disclosure, not all of them. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present disclosure.

[0023] With the continuous development of the gaming industry, online gaming has become an integral part of people's lives. In some highly customizable games, players can freely create their own personalized game scenes (also known as "maps") and interact and communicate with other players. However, due to the freedom and openness of games, some players may behave poorly, using the various components provided to construct structures inappropriate for the content, resulting in a degradation of the gaming environment. Therefore, a solution for monitoring and processing the content of these customized game scenes is needed to maintain a healthy gaming environment and order.

[0024] The complexity and diversity of custom gameplay is a major factor contributing to the difficulty of monitoring. In custom gameplay, players can freely create and design game scenes, rules, props, characters, and other content. The diversity and creativity of this content makes monitoring extremely difficult. Because of this high degree of freedom and creativity, custom gameplay inevitably carries the risk of irregular behavior and inappropriate content, such as vulgarity, violence, and insults.

[0025] However, due to its complexity and diversity, monitoring is extremely difficult, making it difficult to monitor all custom content. Therefore, monitoring and detecting content related to custom gameplay has always been a challenge for the gaming industry. Currently, mainstream products rely on a full manual review process for content monitoring. This involves manually reviewing all game scenes uploaded by players, manually checking for inappropriate content within the scenes, and only releasing them to other players after they have passed the review. While this ensures review quality and avoids misjudgments and missed detections, it also presents challenges such as low review efficiency, high labor consumption, and a negative impact on the player experience.

[0026] First, this type of human review solution is time-consuming and inefficient, making it difficult to meet the review requirements for games with large daily active users (DAU). Due to the sheer volume of custom content, traditional monitoring methods are difficult to effectively monitor. Furthermore, as the number of custom game scenarios increases, manual review becomes even more inefficient.

[0027] Secondly, manually reviewing all game scenes to identify inappropriate content is extremely labor-intensive. The review time for a single game scene exceeds 10 minutes, and a large number of game scenes will result in even greater labor consumption. This manual review method not only requires a large amount of human resources, but is also very time-consuming.

[0028] Finally, the long review process severely impacts the gaming experience for players who choose custom gameplay. A single game scene can take upwards of 10 minutes to review, and as the number of game scenes increases, the review process can be extended to an incalculable amount, severely impacting the player experience.

[0029] In summary, content monitoring for custom gameplay is a challenging task, requiring a combination of manual review and technical means for effective monitoring and processing. We should approach this from all angles, employing a variety of methods to improve review efficiency and accuracy while ensuring a positive gaming experience for players.

[0030] Based on this, the embodiments of the present disclosure provide a method, device, and electronic device for detecting content in a game scene. This technology can be applied to various game scene content detection scenarios.

[0031] Referring to FIG1 , a content detection method for a game scene provided by an embodiment of the present disclosure is first introduced. The method includes the following steps:

[0032] Step S102: Acquire an image to be detected that displays a target game scene.

[0033] Typically, after a game account has created a target game scene and uploaded it to the game system, an image of the target game scene to be detected can be obtained during the review of the target game scene. Alternatively, when a controlled virtual object corresponding to the game account enters the target game scene to play, and the player controlling the game account discovers that the target game scene contains inappropriate content, and takes a screenshot of the current scene, an image of the target game scene to be detected can be obtained.

[0034] When acquiring an image to be inspected that displays a target game scene, the player typically takes a screenshot of the scene using their terminal device. Since this scene is typically captured by a pre-configured virtual camera, the image to be inspected can also be considered to be captured by the virtual camera. For the same target game scene, multiple images to be inspected are typically acquired to perform detailed inspections on different parts of the scene.

[0035] Step S104: input the image to be detected into a pre-trained scene classification model, and process the image to be detected by the scene classification model to obtain a processing result.

[0036] The scene classification model described above is essentially a binary classification model. Images containing the specified content are classified into one category after being processed by the scene classification model, while images not containing the specified content are classified into another category after being processed by the scene classification model. There are many neural networks and deep learning methods for binary image classification, and the specific one can be selected based on your needs.

[0037] The training process for scene classification models typically requires a large number of positive and negative samples. These samples are sample images containing the specified content and sample images excluding the specified content, respectively. Sample images containing the specified content can be manually generated or screenshots of game scene models containing the specified content, as determined during historical review. Furthermore, the accuracy of the scene classification model can be further improved using a contrastive learning loss function, enabling it to accurately classify the images being tested.

[0038] The model usually includes at least a feature extraction module and a fully connected layer module, so that the image features of the image to be detected are extracted through the feature extraction module, and classification is performed based on the image features through the fully connected layer module. Since the target game scene is a three-dimensional model, the feature extraction module can focus on extracting spatial features in the image to be detected. Specifically, this can be achieved through a 3D convolutional neural network. It is usually necessary to perform feature extraction on multiple images to be detected to obtain features such as spatial correlation between the various model parts of the target game scene. After feature extraction of the image to be detected, the category label of the image to be detected can be output as the processing result through the fully connected layer module.

[0039] Step S106: Based on the processing result, determine whether the target game scene contains preset designated content.

[0040] The processing results typically include a category label corresponding to the image to be detected. The category label can indicate whether the image to be detected contains preset specified content. For example, the processing results output by the scene classification model can be A labels and B labels. If the processing result is A label, it means that the corresponding image to be detected does not contain the specified content. If the processing result is B label, it means that the corresponding image to be detected does contain the specified content.

[0041] When the category label corresponding to the image to be detected indicates that it contains the specified content, the target game scene is also determined to contain the specified content. Vice versa. It is worth noting that when a target game scene includes multiple images to be detected, the target game scene can only be determined to contain the specified content if the category labels corresponding to all images to be detected indicate that the image to be detected does not contain the specified content.

[0042] The aforementioned method for detecting game scene content involves acquiring an image showing a target game scene; inputting the image into a pre-trained scene classification model, processing the image using the model to obtain a result; and, based on the result, determining whether the target game scene contains pre-defined content. This method improves the efficiency of reviewing custom game scenes in games, reduces labor costs, shortens player waiting time, and enhances the player's gaming experience.

[0043] The following embodiment provides an implementation method for acquiring an image to be detected that displays a target game scene.

[0044] In specific implementations, images to be detected that display the target game scene can be obtained in a variety of situations, such as when a player uploads the target game scene, when a player actively performs content detection, etc. Several of these situations are described below.

[0045] Game accounts capable of creating game scenes typically generate game scene data through scene editing operations. For example, scene components can be dragged within the game editing scene to arrange the game scene. Once the game account confirms the completion of the edit, the corresponding game scene data is generated. After receiving this game scene data, the game server can generate a target game scene based on the game scene data. In this case, the scene layout within the target game scene matches the scene editing operations performed by the game account. At this point, a preset virtual camera can capture the target game scene at a specified shooting angle to obtain an image to be detected that displays the target game scene. The specified shooting angles include one or more of the following: front, back, left, right, top, and specified tilt angles corresponding to the target game scene. In a specific implementation, the game scene model can be viewed as a cube. The virtual camera captures the scene from the front, back, left, right, and top, as well as at 45° angles relative to the front, back, left, and right sides of the cube, outputting a total of nine images to be detected.

[0046] It is also possible to obtain the image to be detected of the target game scene when the player is trying out the target game scene. Specifically, the game accounts entering the target game scene can be monitored. When a game account accesses the target game scene and the length of stay in the game scene reaches a specified time, a screenshot corresponding to the scene position where the fourth game account is located is obtained, a screenshot is generated, and the screenshot is determined as the image to be detected of the target game scene. The above-mentioned specified time can be multiple, such as 10s, 20s, 1min, etc. After the stay time reaches these specified time, the game scene screens near the scene position where the fourth game account is located are respectively screenshoted to obtain multiple images to be detected. The above-mentioned fourth game account can be a specified user or one or more random users, and can be set according to needs.

[0047] It is also possible to obtain the image to be detected of the target game scene when the game triggers a content detection event. The terminal device used by the player provides a graphical user interface; the graphical user interface displays a scene screen containing the target game scene, and the player can control the controlled virtual object corresponding to the game account used in the target game scene. When the player finds that the current game scene contains specified content, the scene screen can be screenshoted, and the terminal device generates a screenshot. At the same time, the player can also trigger a content detection event by clicking a pre-set content detection control or inputting the screenshot into a pre-set content detection input window. At this time, the screenshot is determined as the image to be detected of the target game scene.

[0048] In addition, the chat messages sent by the first game account can also be monitored; the chat messages sent by the first game account usually include image information or text information; if the chat message includes image information, or the chat message includes text information and the text information includes specified characters, the scene screen is screenshotted to generate a screenshot screen, and the screenshot screen is determined as the image to be detected of the target game scene. The above-mentioned specified characters are usually pre-set words related to the specified content. When the chat message sent by the player includes these words, it can be considered that the player has seen the specified content in the graphical user interface. At this time, a screenshot can be taken and the screenshot can be determined as the image to be detected of the target game scene. When the player sees the specified content in the graphical user interface, he or she may send the screenshot to others for them to view. At this time, a screenshot can also be taken and the screenshot can be determined as the image to be detected of the target game scene.

[0049] In specific implementation, the above scene classification model is trained in the following way;

[0050] 1. Establish an initial model. This initial model can be built based on neural networks, deep learning techniques, etc. The initial model typically includes a basic feature extraction module, a three-dimensional convolutional neural network (3D CNN) module, and a classification module, connected in sequence, as shown in Figure 2.

[0051] 2. Determine training data from a preset sample set; the sample set includes positive samples and negative samples; the positive sample includes a first sample image and a corresponding first category label; the first category label indicates that the first sample image does not contain a preset specified content; the negative sample includes a second sample image and a corresponding second category label; the second category label indicates that the second sample image contains the specified content. In short, the sample images in the positive sample do not contain the preset specified content, while the sample images in the negative sample include the training data, including the preset specified content. The training data determined from the sample set can include either positive samples or negative samples.

[0052] 3. Input the sample images in the training data into the initial model, process the sample images through the initial model, and obtain the first processing result. Specifically, when processing the sample images through the scene classification model, it is necessary to extract the basic image features of the sample images through the basic feature extraction module; extract the three-dimensional scene features of the sample images based on the basic image features through the three-dimensional convolution module; wherein, the three-dimensional scene features include optical flow features, behavioral features and spatial color features, and correspondingly, the three-dimensional convolution module usually includes an optical flow feature extraction module, a behavioral feature extraction module and a spatial color feature extraction module, and the three modules are respectively connected to the basic feature extraction module and the classification module. After obtaining the three-dimensional scene features, the sample images can be classified based on the three-dimensional scene features through the classification model to obtain the category label corresponding to the sample image, and the category label is determined as the first processing result corresponding to the sample image.

[0053] 4. Calculate the loss value of the initial model based on the preset loss function, the first processing result, and the category label corresponding to the sample image. The above loss function can use the common loss function of contrastive learning, and the specific loss function can be selected based on the needs.

[0054] 5. Based on the loss value, adjust the network parameters of the initial model.

[0055] 6. Determine whether the loss value has converged. If not, continue with step 2 above. If converged, determine the trained initial model as the scene classification model.

[0056] After subsequent use, the basic feature extraction module of the scene classification model can extract basic image features of the image to be detected; the three-dimensional scene features of the image to be detected can be extracted based on the basic image features using the three-dimensional convolution module; the three-dimensional scene features include optical flow features, behavioral features, and spatial color features; the classification model can determine the category label corresponding to the image to be detected based on the three-dimensional scene features, and determine the category label as the processing result corresponding to the image to be detected. Furthermore, whether the category label indicates that the image to be detected includes specified content can be further determined to determine whether the target game scene includes the specified content.

[0057] The following embodiment provides an implementation method for determining whether the target game scene has inappropriate content based on the processing results.

[0058] The processing result obtained by processing the scene classification model usually includes a category label corresponding to the image to be detected. The category label is used to indicate whether the image to be detected includes the preset specified content. In a specific implementation, the image to be detected may include multiple. In response to the category label corresponding to at least one of the multiple images to be detected of the target game scene indicating that the image to be detected contains the preset specified content, it is determined that the target game scene model contains the preset specified content; in response to the category labels corresponding to the multiple images to be detected of the target game scene all indicating that the image to be detected does not contain the preset specified content, it is determined that the target game scene model does not contain the preset specified content. Specifically, if one image to be detected is detected to include the specified content, it can be determined that the target game scene includes the specified content.

[0059] The disclosed embodiment also provides another method for detecting content of a game scene, which is implemented based on the method shown in FIG1 .

[0060] To address the aforementioned challenges of difficult detection and high review costs for inappropriate content in custom game scenarios within user-generated content (UGC), a new UGC content detection mechanism and a matching processing model and platform are needed. This standardized process will ensure high coverage of inappropriate content while also being highly efficient and cost-effective, addressing the pain points of UGC custom game scenarios, such as low efficiency, high labor consumption, and negatively impacted player experience.

[0061] The full-process logic and standards for addressing inappropriate content in custom UGC game scenes, from discovery to handling, prioritize the discovery and handling of inappropriate game scenes. Solutions for discovering inappropriate content can include four key areas: proactive game discovery, player screenshot reporting, player-related behavior triggering, and manual image review. Key areas for addressing inappropriate content include training a 3D image detection model (also known as a "3D custom content detection model" or "3D stereoscopic image model") for game scene UGC to improve the accuracy of machine review and reduce risk. The development of a corresponding human review platform, specifically designed for the review of 3D game scenes, including relevant information for human review of game scenes, is also necessary. The mechanism for discovering inappropriate content in custom game scenes is shown in Figure 3, and these components are described in detail below.

[0062] 1. Content is not suitable for discovery mechanism

[0063] 1.1 Active In-Game Discovery: Active in-game discovery refers to setting up mechanisms within the game to automatically detect game scenes that may contain inappropriate content by monitoring the game's operation and player behavior. This mechanism mainly includes two aspects: automatic scanning, and automatic monitoring of screenshots combined with player behavior.

[0064] 1.1.1 Automatic scanning of game scenes: A game map corresponds to a three-dimensional scene. For the three-dimensional scene, there are multiple views from multiple angles (such as a cube).

[0065] Therefore, based on this idea, the game scene can be regarded as a cube, and screenshots can be taken from the front, back, left, right, top, and various angles of the plane angles of the cube. A single game scene can output a total of 9 view screenshots. The view screenshots will output the situation of the game scene from the overall perspective of the game scene. Reviewing the corresponding pictures can solve the problem of discovering inappropriate content in the game scene from a comprehensive perspective.

[0066] 1.1.2 Automatic monitoring of screenshots and gameplay status of game scenes: There are two scenarios for gameplay of a game scene: First, after the game scene is created, the player needs to complete the game scene by themselves before uploading it. This type of gameplay will only appear once for each game scene; second, other players play game scenes created by other authors. In this type of gameplay, each game scene will appear between 10 and tens of millions of times.

[0067] Therefore, we can use the players' gaming habits and situations to set up relevant automatic screenshot trigger nodes:

[0068] ① When the game scene author publishes the scene and offers a trial play, the backend will take screenshots at 30 seconds, 1 minute, and 2 minutes. Gameplay screenshots are generated simultaneously with the scene release and submitted for review. Different user scenarios may not require a trial play, such as a survival game scene. In this case, the above screenshot processing method is not necessary. Alternatively, if the trial play time is insufficient, resulting in a small number of screenshots, the screenshots can be directly used as the images to be tested.

[0069] ② For a game scene that's already in normal play, if the number of players in that scene satisfies the following conditions: "1000 < Number of Players" < 10,000, there's a 0.2% chance that a screenshot will be enabled during the current session. For example, starting at the 30th second, a screenshot will be taken every 30 seconds, and the screenshots will be submitted for review. The upper and lower limits for the number of players and the random probability can be easily adjusted using an online table, enhancing the customization and adjustability of the relevant review logic.

[0070] At the same time, a game scene often has its core gameplay and core content features. Game scene authors often take screenshots of the core content gameplay and use them as the game scene cover. At the same time, the cover image will also be fully reviewed.

[0071] 1.2 Player Screenshot Reporting: This is when a player encounters an in-game scene containing inappropriate content, takes a screenshot, and reports it to the game administrator for review. The greatest advantage of this mechanism is that it allows players to participate in game management, increasing the social nature of the game and player engagement.

[0072] Under this mechanism, the game has set up a reporting system so that players can easily report game scenes with inappropriate content. At the same time, the game administrator also needs to set up an audit system to receive player reports and review the reported content to determine whether there is inappropriate content in the game scene.

[0073] 1.3 Player-related behavior triggers: In terms of automatic content scanning in game scenes, we also conduct in-depth research on player behavior: that is, what kind of behavioral operations do players often perform in the game when they discover inappropriate content in the game scene.

[0074] After analysis, the following two behaviors are common operations performed by players after discovering game scenes with inappropriate content: one is taking screenshots after discovering inappropriate content, which usually includes using in-game screenshot tools and the screenshot button on the mobile phone to take screenshots; the other is sharing the current player's feelings about discovering inappropriate game scenes in text and voice after discovering inappropriate content and discussing it. In this case, the function of blocking inappropriate sensitive words in content is likely to be triggered.

[0075] Therefore, after players discover the above two types of inappropriate content, they are most likely to take actions such as automatic screenshots and review the relevant pictures.

[0076] The advantage of this mechanism is that it can promptly detect game scenes that may contain inappropriate content, but it also has certain flaws, because this mechanism can only detect game scenes with inappropriate content related to player behavior, and cannot detect game scenes with inappropriate content but not triggered by player behavior.

[0077] 1.4 Manual Gameplay Discovery: Manual gameplay discovery involves game administrators manually running gameplay to identify potentially inappropriate game scenes. While this mechanism is cumbersome, it is currently one of the most commonly used methods. Under this mechanism, administrators must meticulously inspect all aspects of the game scene, such as its size, the number of resources used, its naming, its labeling, and so on. This type of audit is complex and challenging, so improving the efficiency of audit staff requires developing a custom platform to assist with audits. This section will be described later.

[0078] II. Mechanism for Handling Inappropriate Content

[0079] 2.1 Establishing a new 3D custom content detection model

[0080] 3D Content Detection Model Overview

[0081] A 3D model is a computer-generated three-dimensional graphic composed of a series of points, lines, and surfaces. In game development, 3D models can be used to create various in-game objects, such as characters and scenes. A 3D model is typically composed of multiple triangular faces, each of which consists of three vertices, each of which contains information such as coordinates and color.

[0082] 2.2 3D Content Detection Model Training Scheme

[0083] The training scheme for the 3D content detection model can be divided into the following steps:

[0084] A. Data Preprocessing

[0085] Data preprocessing is a crucial step for 3D content detection models. Since 3D image models require a large amount of data, data compression and cleaning are necessary to facilitate subsequent training. Data preprocessing specifically includes the following aspects:

[0086] (1) Data compression: The data in the 3D stereo image model is compressed through a compression algorithm to reduce the size of the data and improve training efficiency.

[0087] (2) Data cleaning: Clean out useless information such as noise and impurities in the 3D stereo image model to improve the accuracy of the training model.

[0088] (3) Data standardization: The data in the 3D stereo image model is standardized to meet the input requirements of the training model.

[0089] B. Data Augmentation

[0090] (1) Single sample data enhancement: perform geometric and color space transformations.

[0091] (2) Multi-sample data enhancement: SMOTE, mixup, and mosaic data enhancement methods are used.

[0092] (3) Unsupervised data enhancement: using GAN to generate adversarial samples.

[0093] C. Model Construction

[0094] After data preprocessing is completed, the model needs to be built. Model building refers to building a 3D stereo image model training model based on the training data. The model building process includes the following aspects:

[0095] (1) Determine the training algorithm: According to the characteristics of the training data, select a suitable training algorithm and build a classification network based on a deep convolutional network.

[0096] (2) Determine the structure of the training model: According to the characteristics of the training data, determine the structure of the training model and add 3D convolution to improve the stereo image recognition performance.

[0097] (3) Determine the training parameters: According to the characteristics of the training data, determine the parameters of the training algorithm, introduce 3D image multi-frame training, introduce contrastive learning and adversarial loss.

[0098] D. Training Model

[0099] After the model is built, it needs to be trained. Training the model refers to the process of inputting training data into the model to learn and optimize the model. The process of training the model includes the following aspects:

[0100] (1) Dataset division: The training dataset is divided into training set, validation set and test set for training the model, validating the model and testing the accuracy of the model.

[0101] (2) Parameter initialization: Initialize the training parameters to random values.

[0102] (3) Training the model: Input the training data into the model, and learn and optimize the model until the model's training error reaches a certain threshold. As shown in Figure 4, the model can separate positive and negative samples. After training, the model can accurately separate government samples. The samples above the dotted line are negative samples, and the samples below the dotted line are positive samples.

[0103] (4) Model evaluation: Use the test set to evaluate the model and calculate the model's accuracy, recall rate, F1 value and other indicators.

[0104] (5) Model optimization: Based on the model evaluation results, the model is optimized, including adjusting training parameters, modifying the model structure, etc., using distributed training, dropout, early stopping, labelsmoothing, Adam optimization, etc.

[0105] E. Model Prediction

[0106] After model training is complete, the model needs to be used to predict new 3D stereo image models. Model prediction involves inputting a new 3D stereo image model into the model and predicting its category. The model prediction process includes the following aspects:

[0107] (1) Data preprocessing: Preprocess the new 3D stereo image model to make it meet the input requirements of the training model.

[0108] (2) Model prediction: Input the new 3D stereo image model into the trained model to predict the category it belongs to.

[0109] (3) Result output: Output the prediction results to a file or database for subsequent application.

[0110] After the new model training was completed, the monthly call count was over 20 million, and the model recognition effect was shown in Table 1:

[0111] Table 1

[0112] 3. Establish a dedicated manual review platform for game scene content review

[0113] In addition to establishing a custom review model, manual review is still required for some key game scenes. However, the conventional manual review process is complex and difficult to operate, involves many systems and permissions, and it is difficult to display the full amount of information about the game scene. It is necessary to establish a manual review platform dedicated to game scene review. This platform should include the following aspects:

[0114] 3.1 Basic Architecture of the Audit Platform

[0115] The basic architecture of the audit platform is divided into two parts: the front-end and the back-end. The front-end is mainly responsible for user input and display, while the back-end is mainly responsible for data processing and storage.

[0116] The front-end includes web pages and mobile applications, providing players with functions such as uploading game scenes and querying review status. The back-end includes databases and servers, and is mainly responsible for the management, storage, and processing of game scene review data.

[0117] 3.2 All information on game scene review

[0118] 3.2.1 Game scene upload information

[0119] The game scene upload information includes the game scene name, game scene description, game scene author, game scene type, upload time, etc. This information is used to identify and manage game scenes.

[0120] 3.2.2 Game scene content information

[0121] Game scene content information includes game scene size, game scene difficulty, game scene pattern, game scene gameplay, etc. This information is used to review whether the game scene complies with regulations.

[0122] 3.2.3 Game scene review status information

[0123] Game scene review status information includes game scene review status, review time, reviewer, etc. This information is used to record the process and results of the game scene review.

[0124] 3.2.4 Game scene review opinion information

[0125] Game scene review opinion information includes review suggestions, reasons for passing / failure of the review, etc. This information is used to provide review results and improvement suggestions to game scene authors.

[0126] 3.3 Related interfaces involved

[0127] 3.3.1 Game scene upload interface

[0128] The game scene upload interface is used for players to upload their own game scenes.

[0129] 3.3.2 Game scene query interface

[0130] The game scene query interface is used to query uploaded game scene information. Players can search for game scenes by keywords.

[0131] 3.3.3 Game scene review interface

[0132] The game scene review interface is used to review game scenes. Reviewers can review game scenes and provide review results and suggestions.

[0133] 3.3.4 Game scene deletion interface

[0134] The game scene deletion interface is used to delete game scenes that do not meet the requirements.

[0135] 3.3.5 Game scene editing interface

[0136] The game scene editing interface is used to edit uploaded game scenes. Game scene authors can modify the game scene content before review.

[0137] 3.3.6 Game scene synchronization interface

[0138] The game scene download interface is used to download approved game scenes. Players can download their favorite game scenes.

[0139] 3.3.7 Statistics Interface

[0140] The statistical interface is used to count data such as the number of game scenes that have been reviewed and approved, the number of game scene downloads, etc., in order to evaluate the effectiveness of the review platform and the quality of game scenes.

[0141] 3.4 Operation and maintenance of the audit platform

[0142] 3.4.1 Operations

[0143] The operation of the review platform requires regular promotion and publicity to attract more players to upload and download game scenes. At the same time, it is necessary to manage the review personnel well to ensure the quality and efficiency of the review.

[0144] 3.4.2 Maintenance

[0145] The audit platform needs to be regularly updated and upgraded to ensure its stability and security. At the same time, data needs to be backed up regularly to prevent data loss.

[0146] 3.5 Summary

[0147] The proposed platform for manually reviewing custom game scenes aims to improve the efficiency and quality of scene reviews and ensure a positive gaming experience for players. The platform's basic architecture consists of both front-end and back-end components. All scene review information includes scene upload information, scene content, scene review status, and scene review opinions. The platform's eight game interfaces include scene upload, scene query, scene review, scene deletion, scene editing, scene download, scene sharing, and statistics. The platform manages review personnel and regularly updates and upgrades the platform.

[0148] In short, establishing a new UGC custom game scene management process involves both identifying and handling inappropriate content. Mechanisms for identifying inappropriate content include proactive game discovery, player screenshot reporting, player-related behavior triggering, and manual game discovery. The mechanism for handling inappropriate content includes establishing a new 3D custom content detection model and a dedicated manual review platform for game scenes. Establishing this process is crucial to ensuring the quality of game scenes and the healthy development of the game.

[0149] In the above method, in order to solve the many problems existing in the solution relying on full manual review, the game scene screenshot review method is adopted, including three aspects: active discovery of the game, discovery through player screenshot reports, and triggering of player-related behaviors. 6.87 million game scenes are reviewed monthly, and only 1,500 game scenes are reviewed manually. Automatic review accounts for 99.8%, which greatly reduces the cost investment of manual review by 99.8%. At the same time, it improves the player game scene approval time to within 3 seconds, which is 99.5% lower than the efficiency of manual review of game scenes of 10 minutes.

[0150] Regarding the 3D image detection model for gaming UGC, conventional image models are not perfectly adapted to the content in 3D gaming scenes, resulting in frequent missed detections. Therefore, a dedicated model was developed specifically for these scenarios. This model was trained using relevant in-game images, resulting in an improvement in its ability to identify inappropriate 3D content images from 60% to 96%.

[0151] Regarding the human-reviewed game scene platform, it aggregates eight interfaces, including query processing, and is connected to the game itself. This allows reviewers to complete the review of relevant game scenes in the shortest possible time, improving efficiency.

[0152] Regarding the above method embodiment, referring to FIG5 , an embodiment of the present disclosure provides a content detection device for a game scene, the device comprising:

[0153] The image acquisition module 502 is configured to acquire an image to be detected that displays a target game scene;

[0154] The classification module 504 is configured to input the image to be detected into a pre-trained scene classification model, process the image to be detected by the scene classification model, and obtain a processing result;

[0155] The content determination module 506 is configured to determine whether the target game scene contains preset specified content based on the processing result.

[0156] The aforementioned game scene content detection device captures an image showing a target game scene; inputs the image into a pre-trained scene classification model, processes the image using the model, and obtains a processing result; based on the processing result, determines whether the target game scene contains pre-set, specified content. This method improves the efficiency of custom game scene review in games, reduces labor costs, shortens player waiting time, and enhances the player's gaming experience.

[0157] The above-mentioned image acquisition module to be detected is also configured to execute: obtaining game scene data; the game scene data is generated based on the scene editing operation corresponding to the first game account; based on the game scene data, a target game scene is generated; the target game scene is photographed by a preset virtual camera based on a specified shooting angle to obtain an image to be detected showing the target game scene; the specified shooting angle includes one or more of the front angle, back angle, left angle, right angle, top angle and specified tilt angle corresponding to the target game scene.

[0158] The above method provides a graphical user interface through a terminal device; the graphical user interface displays a scene screen corresponding to the target game scene; the above-mentioned image acquisition module is further configured to execute: obtaining a screenshot corresponding to the scene screen generated by the second game account when a content detection event is triggered, and determining the screenshot as the image to be detected of the target game scene.

[0159] The above method provides a graphical user interface through a terminal device; the graphical user interface displays a scene screen corresponding to the target game scene; the above-mentioned image acquisition module to be detected is further configured to execute: monitoring chat information sent by a third game account; the chat information includes image information or text information; the third game account is controlled by the terminal device; if the chat information includes image information, or the chat information includes text information and the text information includes specified characters, the scene screen is screenshoted to generate a screenshot screen, and the screenshot screen is determined as the image to be detected of the target game scene.

[0160] The above-mentioned image acquisition module to be detected is also configured to execute: monitoring the game accounts entering the target game scene; if the fourth game account accesses the target game scene and the length of stay in the game scene reaches a specified time, obtaining a screenshot corresponding to the scene location where the fourth game account is located, generating a screenshot, and determining the screenshot as the image to be detected of the target game scene model.

[0161] In the above-mentioned device, a graphical user interface is provided through the terminal device; the graphical user interface displays a scene screen containing the target game scene; the above-mentioned image acquisition module to be detected is also configured to execute: generating a screenshot screen in response to a screenshot operation on the scene screen; determining the screenshot screen as the image to be detected of the target game scene in response to the triggering operation of the content detection event.

[0162] In the above-mentioned device, a graphical user interface is provided through the terminal device; the graphical user interface displays a scene screen generated by a preset virtual camera shooting a target game scene; the above-mentioned image acquisition module to be detected is further configured to execute: monitoring chat information sent by a first game account; the chat information includes image information or text information; the first game account is controlled by the terminal device; in response to the chat information including image information, or the chat information including text information and the text information including specified characters, the scene screen is screenshoted to generate a screenshot screen, and the screenshot screen is determined as the image to be detected of the target game scene.

[0163] In the above-mentioned device, a graphical user interface is provided through the terminal device; the above-mentioned image to be detected acquisition module is further configured to execute: in response to the controlled virtual object corresponding to the first game account entering the game scene corresponding to the target game scene, the scene screen corresponding to the target game scene is displayed on the graphical user interface; the controlled virtual object is controlled by the terminal device; in response to the controlled virtual object staying in the game scene for a specified time, the scene screen is screenshotted to generate a screenshot, and the screenshot is determined as the image to be detected of the target game scene.

[0164] The above-mentioned device also includes a model training module, which is configured to execute: establish an initial model; determine training data from a preset sample set; the sample set includes positive samples and negative samples; the positive sample includes a first sample image and a corresponding first category label; the first category label indicates that the first sample image does not contain a preset specified content; the negative sample includes a second sample image and a corresponding second category label; the second category label indicates that the second sample image contains specified content; the training data includes positive samples or negative samples; input the sample image in the training data into the initial model, and process the sample image through the initial model to obtain a first processing result; calculate the loss value of the initial model based on the preset loss function, the first processing result and the category label corresponding to the sample image; adjust the network parameters of the initial model based on the loss value; continue to execute the step of determining training data from the preset sample set until the loss value converges, and determine the trained initial model as the scene classification model.

[0165] The above-mentioned scene classification model includes a basic feature extraction module, a three-dimensional convolution module and a classification module connected in sequence; the above-mentioned classification module is also configured to perform: extracting the basic image features of the image to be detected through the basic feature extraction module; extracting the three-dimensional scene features of the image to be detected based on the basic image features through the three-dimensional convolution module; the three-dimensional scene features include optical flow features, behavioral features and spatial color features; determining the category label corresponding to the image to be detected based on the three-dimensional scene features through the classification model, and determining the category label as the processing result corresponding to the image to be detected.

[0166] The above processing results include a category label corresponding to the image to be detected; the category label is used to indicate whether the image to be detected includes a preset specified content; the image to be detected includes multiple images; the above content determination module is also configured to execute: in response to the category label corresponding to at least one of the multiple images to be detected of the target game scene indicating that the image to be detected contains the preset specified content, determining that the target game scene model contains the preset specified content; in response to the category labels corresponding to the multiple images to be detected of the target game scene all indicating that the image to be detected does not contain the preset specified content, determining that the target game scene model does not contain the preset specified content.

[0167] This embodiment also provides an electronic device, including a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor executes the machine-executable instructions to implement the above-mentioned game scene content detection method. The details are as follows:

[0168] Obtain an image to be detected that displays a target game scene; input the image to be detected into a pre-trained scene classification model, process the image to be detected through the scene classification model to obtain a processing result; based on the processing result, determine whether the target game scene contains preset specified content.

[0169] The above method improves the efficiency of reviewing custom game scenes in the game, reduces labor costs, shortens players' waiting time, and improves players' gaming experience.

[0170] Optionally, the above-mentioned step of obtaining the image to be detected showing the target game scene includes: obtaining game scene data; the game scene data is generated based on the scene editing operation corresponding to the first game account; based on the game scene data, the target game scene is generated; the target game scene is photographed by a preset virtual camera based on a specified shooting angle to obtain the image to be detected showing the target game scene; the specified shooting angle includes one or more of the front angle, back angle, left angle, right angle, top angle and specified tilt angle corresponding to the target game scene.

[0171] Optionally, the above method provides a graphical user interface through a terminal device; the graphical user interface displays a scene screen corresponding to the target game scene; the step of obtaining an image to be detected that displays the target game scene includes: obtaining a screenshot corresponding to the scene screen generated by the second game account when a content detection event is triggered, and determining the screenshot as the image to be detected of the target game scene.

[0172] Optionally, the above method provides a graphical user interface through the terminal device; the graphical user interface displays a scene screen corresponding to the target game scene; the step of obtaining the image to be detected that displays the target game scene includes: monitoring chat information sent by a third game account; the chat information includes image information or text information; the third game account is controlled by the terminal device; if the chat information includes image information, or the chat information includes text information and the text information includes specified characters, the scene screen is screenshoted to generate a screenshot, and the screenshot is determined as the image to be detected of the target game scene.

[0173] Optionally, the above-mentioned step of obtaining the image to be detected that displays the target game scene model includes: monitoring the game accounts entering the target game scene; if the fourth game account accesses the target game scene and the length of stay in the game scene reaches a specified time, obtaining a screenshot corresponding to the scene position where the fourth game account is located, generating a screenshot, and determining the screenshot as the image to be detected of the target game scene model.

[0174] Optionally, the above-mentioned scene classification model is trained in the following manner: establishing an initial model; determining training data from a preset sample set; the sample set includes positive samples and negative samples; the positive sample includes a first sample image and a corresponding first category label; the first category label indicates that the first sample image does not contain a preset specified content; the negative sample includes a second sample image and a corresponding second category label; the second category label indicates that the second sample image contains specified content; the training data includes positive samples or negative samples; inputting the sample image in the training data into the initial model, processing the sample image through the initial model to obtain a first processing result; calculating the loss value of the initial model based on the preset loss function, the first processing result and the category label corresponding to the sample image; adjusting the network parameters of the initial model based on the loss value; continuing to execute the step of determining training data from the preset sample set until the loss value converges, and determining the trained initial model as the scene classification model.

[0175] Optionally, the above-mentioned scene classification model includes a basic feature extraction module, a three-dimensional convolution module and a classification module connected in sequence; the step of processing the image to be detected through the scene classification model to obtain the processing result includes: extracting the basic image features of the image to be detected through the basic feature extraction module; extracting the three-dimensional scene features of the image to be detected based on the basic image features through the three-dimensional convolution module; the three-dimensional scene features include optical flow features, behavioral features and spatial color features; determining the category label corresponding to the image to be detected based on the three-dimensional scene features through the classification model, and determining the category label as the processing result corresponding to the image to be detected.

[0176] Optionally, the above-mentioned processing results include a category label corresponding to the image to be detected; the category label is used to indicate whether the image to be detected includes preset specified content; the image to be detected includes multiple images; based on the processing results, the step of determining whether the content of the target game scene is inappropriate includes: in response to the category label corresponding to at least one of the multiple images to be detected of the target game scene indicating that the image to be detected contains the preset specified content, determining that the target game scene model contains the preset specified content; in response to the category labels corresponding to the multiple images to be detected of the target game scene all indicating that the image to be detected does not contain the preset specified content, determining that the target game scene model does not contain the preset specified content.

[0177] As shown in FIG6 , the electronic device includes a processor 100 and a memory 101 . The memory 101 stores machine-executable instructions that can be executed by the processor 100 . The processor 100 executes the machine-executable instructions to implement the above-mentioned content detection method for the game scene.

[0178] Furthermore, the electronic device shown in FIG6 further includes a bus 102 and a communication interface 103 , and the processor 100 , the communication interface 103 and the memory 101 are connected via the bus 102 .

[0179] Memory 101 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between the system network element and at least one other network element is achieved through at least one communication interface 103 (which may be wired or wireless), and may utilize the Internet, a wide area network, a local area network, a metropolitan area network, etc. Bus 102 may be an ISA bus, a PCI bus, or an EISA bus. The bus may be classified as an address bus, a data bus, a control bus, etc. For ease of illustration, FIG6 shows only one bidirectional arrow, but this does not imply that there is only one bus or only one type of bus.

[0180] The processor 100 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by an integrated logic circuit of hardware in the processor 100 or by instructions in the form of software. The above-mentioned processor 100 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, or discrete hardware components. The various methods, steps, and logic block diagrams disclosed in the embodiments of the present disclosure can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present disclosure can be directly embodied as being executed by a hardware decoding processor, or can be executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium known in the art, such as a random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), electrically erasable programmable memory (EEPM), or registers. The storage medium is located in memory 101. Processor 100 reads information from memory 101 and, in conjunction with its hardware, performs the steps of the method of the aforementioned embodiment.

[0181] This embodiment also provides a machine-readable storage medium, which stores machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions prompt the processor to implement the above-mentioned game scene content detection method.

[0182] The present disclosure provides a method, device, and electronic device for detecting content in a game scene, including a computer-readable storage medium storing program code. The program code includes instructions that can be used to execute the methods described in the previous method embodiments. Specifically,

[0183] Obtain an image to be detected that displays a target game scene; input the image to be detected into a pre-trained scene classification model, process the image to be detected through the scene classification model to obtain a processing result; based on the processing result, determine whether the target game scene contains preset specified content.

[0184] The above method improves the efficiency of reviewing custom game scenes in the game, reduces labor costs, shortens players' waiting time, and improves players' gaming experience.

[0185] Optionally, the above-mentioned step of obtaining the image to be detected showing the target game scene includes: obtaining game scene data; the game scene data is generated based on the scene editing operation corresponding to the first game account; based on the game scene data, the target game scene is generated; the target game scene is photographed by a preset virtual camera based on a specified shooting angle to obtain the image to be detected showing the target game scene; the specified shooting angle includes one or more of the front angle, back angle, left angle, right angle, top angle and specified tilt angle corresponding to the target game scene.

[0186] Optionally, the above method provides a graphical user interface through a terminal device; the graphical user interface displays a scene screen corresponding to the target game scene; the step of obtaining an image to be detected that displays the target game scene includes: obtaining a screenshot corresponding to the scene screen generated by the second game account when a content detection event is triggered, and determining the screenshot as the image to be detected of the target game scene.

[0187] Optionally, the above method provides a graphical user interface through the terminal device; the graphical user interface displays a scene screen corresponding to the target game scene; the step of obtaining the image to be detected that displays the target game scene includes: monitoring chat information sent by a third game account; the chat information includes image information or text information; the third game account is controlled by the terminal device; if the chat information includes image information, or the chat information includes text information and the text information includes specified characters, the scene screen is screenshoted to generate a screenshot, and the screenshot is determined as the image to be detected of the target game scene.

[0188] Optionally, the above-mentioned step of obtaining the image to be detected that displays the target game scene model includes: monitoring the game accounts entering the target game scene; if the fourth game account accesses the target game scene and the length of stay in the game scene reaches a specified time, obtaining a screenshot corresponding to the scene position where the fourth game account is located, generating a screenshot, and determining the screenshot as the image to be detected of the target game scene model.

[0189] Optionally, the above-mentioned scene classification model is trained in the following manner: establishing an initial model; determining training data from a preset sample set; the sample set includes positive samples and negative samples; the positive sample includes a first sample image and a corresponding first category label; the first category label indicates that the first sample image does not contain a preset specified content; the negative sample includes a second sample image and a corresponding second category label; the second category label indicates that the second sample image contains specified content; the training data includes positive samples or negative samples; inputting the sample image in the training data into the initial model, processing the sample image through the initial model to obtain a first processing result; calculating the loss value of the initial model based on the preset loss function, the first processing result and the category label corresponding to the sample image; adjusting the network parameters of the initial model based on the loss value; continuing to execute the step of determining training data from the preset sample set until the loss value converges, and determining the trained initial model as the scene classification model.

[0190] Optionally, the above-mentioned scene classification model includes a basic feature extraction module, a three-dimensional convolution module and a classification module connected in sequence; the step of processing the image to be detected through the scene classification model to obtain the processing result includes: extracting the basic image features of the image to be detected through the basic feature extraction module; extracting the three-dimensional scene features of the image to be detected based on the basic image features through the three-dimensional convolution module; the three-dimensional scene features include optical flow features, behavioral features and spatial color features; determining the category label corresponding to the image to be detected based on the three-dimensional scene features through the classification model, and determining the category label as the processing result corresponding to the image to be detected.

[0191] Optionally, the above-mentioned processing results include a category label corresponding to the image to be detected; the category label is used to indicate whether the image to be detected includes preset specified content; the image to be detected includes multiple images; based on the processing results, the step of determining whether the content of the target game scene is inappropriate includes: in response to the category label corresponding to at least one of the multiple images to be detected of the target game scene indicating that the image to be detected contains the preset specified content, determining that the target game scene model contains the preset specified content; in response to the category labels corresponding to the multiple images to be detected of the target game scene all indicating that the image to be detected does not contain the preset specified content, determining that the target game scene model does not contain the preset specified content.

[0192] As shown in FIG6 , the electronic device includes a processor 100 and a memory 101 . The memory 101 stores machine-executable instructions that can be executed by the processor 100 . The processor 100 executes the machine-executable instructions to implement the above-mentioned content detection method for the game scene.

[0193] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems and devices can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0194] In addition, in the description of the embodiments of the present disclosure, unless otherwise expressly specified or limited, the terms "installed," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in the present disclosure based on the specific circumstances.

[0195] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present disclosure, or the part that contributes to the relevant technology or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0196] In the description of this disclosure, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate the description of this disclosure and simplify the description. They do not indicate or imply that the devices or components referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on this disclosure. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0197] Finally, it should be noted that the above embodiments are only specific implementation methods of the present disclosure, which are used to illustrate the technical solutions of the present disclosure, rather than to limit them. The scope of protection of the present disclosure is not limited thereto. Although the present disclosure has been described in detail with reference to the above embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above embodiments within the technical scope disclosed in the present disclosure, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in the scope of protection of the present disclosure. Therefore, the scope of protection of the present disclosure should be based on the scope of protection of the claims.

Claims

1. A method for detecting content of a game scene, comprising: Acquire an image to be detected showing a target game scene; Inputting the image to be detected into a pre-trained scene classification model, processing the image to be detected by the scene classification model to obtain a processing result; Based on the processing result, it is determined whether the target game scene contains preset designated content.

2. The method according to claim 1, wherein: The step of obtaining an image to be detected showing a target game scene includes: Acquire game scene data; the game scene data is generated based on a scene editing operation corresponding to the first game account; Based on the game scene data, generate a target game scene; The target game scene is photographed by a preset virtual camera based on a specified shooting angle to obtain an image to be detected that displays the target game scene; the specified shooting angle includes one or more of the front angle, back angle, left angle, right angle, top angle and specified tilt angle corresponding to the target game scene.

3. The method according to claim 1, wherein: Providing a graphical user interface through a terminal device; the graphical user interface displays a scene screen corresponding to the target game scene; The step of obtaining an image to be detected showing a target game scene includes: A screenshot corresponding to the scene screen generated by the second game account when a content detection event is triggered is obtained, and the screenshot is determined as the image to be detected of the target game scene.

4. The method according to claim 1, wherein: Providing a graphical user interface through a terminal device; the graphical user interface displays a scene screen corresponding to the target game scene; The step of obtaining an image to be detected showing a target game scene includes: monitoring chat information sent by a third game account; the chat information includes image information or text information; the third game account is controlled by the terminal device; If the chat information includes image information, or the chat information includes text information and the text information includes specified characters, a screenshot is taken of the scene screen to generate a screenshot screen, and the screenshot screen is determined as the image to be detected of the target game scene.

5. The method according to claim 1, wherein: The step of obtaining an image to be detected showing a target game scene model comprises: Monitoring game accounts entering the target game scene; If the fourth game account visits the target game scene and stays in the game scene for a specified time, a screenshot corresponding to the scene location where the fourth game account is located is obtained, a screenshot is generated, and the screenshot is determined as the image to be detected of the target game scene model.

6. The method according to claim 1, wherein: The scene classification model is trained in the following way; Build an initial model; Determine training data from a preset sample set; the sample set includes positive samples and negative samples; the positive samples The method comprises a first sample image and a corresponding first category label; the first category label indicates that the first sample image does not contain a preset specified content; the negative sample comprises a second sample image and a corresponding second category label; the second category label indicates that the second sample image contains the specified content; the training data comprises the positive sample or the negative sample; Inputting a sample image in the training data into the initial model, and processing the sample image by the initial model to obtain a first processing result; Calculating the loss value of the initial model based on a preset loss function, the first processing result, and the category label corresponding to the sample image; Based on the loss value, adjusting the network parameters of the initial model; The step of determining training data from a preset sample set is continued until the loss value converges, and the trained initial model is determined as a scene classification model.

7. The method according to claim 6, wherein: The scene classification model includes a basic feature extraction module, a three-dimensional convolution module and a classification module connected in sequence; The step of processing the image to be detected by using the scene classification model to obtain a processing result includes: Extracting basic image features of the image to be detected by the basic feature extraction module; Extracting three-dimensional scene features of the image to be detected based on the basic image features through the three-dimensional convolution module; the three-dimensional scene features include optical flow features, behavior features and spatial color features; The classification model is used to determine a category label corresponding to the image to be detected based on the three-dimensional scene features, and the category label is determined as a processing result corresponding to the image to be detected.

8. The method according to claim 1, wherein: The processing result includes a category label corresponding to the image to be detected; the category label is used to indicate whether the image to be detected includes a preset specified content; the image to be detected includes multiple images; Based on the processing result, the step of determining whether the target game scene contains preset specified content includes: In response to a category label corresponding to at least one image to be detected among a plurality of images to be detected of the target game scene indicating that the image to be detected contains preset specified content, determining that the target game scene model contains the preset specified content; In response to the category labels corresponding to the multiple images to be detected of the target game scene all indicating that the images to be detected do not contain preset designated content, it is determined that the target game scene model does not contain the preset designated content.

9. A content detection device for a game scene, comprising: The to-be-detected image acquisition module is configured to execute acquisition of the to-be-detected image showing the target game scene; A classification module is configured to input the image to be detected into a pre-trained scene classification model, process the image to be detected by the scene classification model, and obtain a processing result; The content determination module is configured to determine whether the target game scene contains preset designated content based on the processing result.

10. An electronic device comprising a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor executes the machine executable instructions to implement the content detection method of the game scene described in any one of claims 1-8.

11. A machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are called and executed by a processor, the machine-executable instructions prompt the processor to implement the content detection method for a game scene as described in any one of claims 1-8.

Images