Information detection method and device, electronic equipment, computer readable storage medium and computer program product

Abstract

The invention provides an information detection method and device, electronic equipment, a computer readable storage medium and a computer program product. The information detection method and device are applied to various mold penetrating detection scenes such as cloud technology, artificial intelligence, intelligent transportation and games. The information detection method comprises the steps that attention processing is carried out on initial image features of a to-be-detected image, to-be-detected image features are obtained, and the to-be-detected image is an image of a virtual object model; performing through-mold detection on the to-be-detected image features to obtain a first through-mold detection result; based on the image semantic information of the to-be-detected image, an element graph network is constructed, and the element graph network represents a semantic relationship between image elements in the to-be-detected image; performing mold penetrating detection on the element graph network to obtain a second mold penetrating detection result; and combining the first mold penetrating detection result and the second mold penetrating detection result to obtain a target mold penetrating detection result of the to-be-detected image. According to the invention, the accuracy and efficiency of mold penetration detection can be improved.

Description

Technical Field

[0001] This application relates to information detection technology in the field of information processing, and more particularly to an information detection method, apparatus, electronic device, computer-readable storage medium, and computer program product. Background Technology

[0002] Clipping detection is used to detect whether virtual object models exhibit unrealistic clipping phenomena. Therefore, by detecting clipping, it is possible to determine whether virtual object models in a virtual scene can correctly follow the rules of the physical world.

[0003] Generally, clipping detection is usually performed manually; that is, it is determined by humans whether the virtual object model in the virtual scene has clipped. This affects the accuracy and efficiency of clipping detection. Summary of the Invention

[0004] This application provides an information detection method, apparatus, electronic device, computer-readable storage medium, and computer program product, which can improve the accuracy and efficiency of mold penetration detection.

[0005] The technical solution of this application embodiment is implemented as follows:

[0006] This application provides an information detection method, the method comprising:

[0007] Attention processing is performed on the initial image features of the image to be detected to obtain the image features to be detected, wherein the image to be detected is an image of a virtual object model;

[0008] The features of the image to be detected are subjected to clipping detection to obtain a first clipping detection result;

[0009] Based on the image semantic information of the image to be detected, an element graph network is constructed, which represents the semantic relationship between each image element in the image to be detected.

[0010] The element graph network is subjected to clipping detection to obtain a second clipping detection result;

[0011] By combining the first clipping detection result and the second clipping detection result, the target clipping detection result of the image to be detected is obtained.

[0012] This application provides an information detection device, the information detection device comprising:

[0013] The feature processing module is used to perform attention processing on the initial image features of the image to be detected to obtain the image features to be detected, wherein the image to be detected is an image of a virtual object model;

[0014] The first detection module is used to perform clipping detection on the features of the image to be detected, and obtain a first clipping detection result;

[0015] The network construction module is used to construct an element graph network based on the image semantic information of the image to be detected, wherein the element graph network represents the semantic relationship between various image elements in the image to be detected.

[0016] The second detection module is used to perform clipping detection on the element graph network and obtain a second clipping detection result.

[0017] The result acquisition module is used to combine the first clipping detection result and the second clipping detection result to obtain the target clipping detection result of the image to be detected.

[0018] In this embodiment of the application, the feature processing module is further configured to perform attention processing on the initial image features of the image to be detected to obtain the attention weights of the initial image features; and to fuse the initial image features and the attention weights to obtain the image features to be detected.

[0019] In this embodiment of the application, the semantic relationship includes at least one of the following: hierarchical relationship, positional relationship, and action relationship; the hierarchical relationship represents the display hierarchy between the various image elements, the positional relationship represents the relative position between the various image elements, and the action relationship represents the rigid body motion action between the various image elements.

[0020] In this embodiment of the application, the network construction module is further configured to extract the image semantic features of the image to be detected; and determine the image semantic information of the image to be detected based on the image semantic features.

[0021] In this embodiment of the application, the network construction module is further configured to determine each of the image elements of the image to be detected and the semantic relationship between each of the image elements from the image semantic information; and to construct the element graph network by using each of the image elements as graph network nodes and constructing graph network node edges based on the semantic relationship.

[0022] In this embodiment of the application, the feature processing module is further configured to record video of the process of the virtual object model executing motion commands to obtain a video to be detected; divide the video to be detected into video frames to obtain a video frame image sequence; traverse the video frame image sequence and determine the traversed video frame image as the image to be detected.

[0023] In this embodiment of the application, the second detection module is further configured to perform clipping detection on the element graph network to obtain a third clipping detection result; obtain N fourth clipping detection results corresponding to the first N video frames of the image to be detected, where N is a positive integer; and combine the third clipping detection result and the N fourth clipping detection results to obtain the second clipping detection result.

[0024] In this embodiment of the application, the feature processing module is further configured to perform image preprocessing on the image to be detected to obtain an image to be processed, wherein the image preprocessing includes at least one of the following: noise reduction, brightness adjustment and contrast adjustment; and to extract features from the image to be processed to obtain the initial image features.

[0025] In this embodiment of the application, the result acquisition module is further configured to: when both the first and second clipping detection results indicate that the virtual object model is clipping, determine the information in the image to be detected that the virtual object model is clipping as the target clipping detection result; when both the first and second clipping detection results indicate that the virtual object model is not clipping, determine the information in the image to be detected that the virtual object model is not clipping as the target clipping detection result; and when one of the first and second clipping detection results indicates that the virtual object model is clipping, determine the information in the image to be detected that the virtual object model is to be clipped and verified as the target clipping detection result.

[0026] In this embodiment, the first clipping detection result is achieved through a first clipping detection model; the information detection device further includes a model training module for acquiring a first image sample and a first clipping annotation result of the first image sample; performing clipping detection on the first image sample using a first model to be trained to obtain a first clipping prediction result, wherein the first model to be trained is a neural network model to be trained for clipping detection based on attention processing; and training the first model to be trained based on the difference between the first clipping prediction result and the first clipping annotation result to obtain the first clipping detection model.

[0027] In this embodiment, the second clipping detection result is achieved through a second clipping detection model; the model training module is used to obtain a second image sample and a second clipping annotation result of the second image sample; to perform clipping detection on the second image sample using a second model to be trained, and to obtain a second clipping prediction result, wherein the second model to be trained is a neural network model to be trained for clipping detection based on semantic information; and to train the second model to be trained based on the difference between the second clipping prediction result and the second clipping annotation result, thereby obtaining the second clipping detection model.

[0028] In this embodiment of the application, the information detection device further includes a model reconstruction module, which is used to reconstruct the virtual object model based on the image to be detected when the target clipping detection result indicates that the virtual object model clips through the image to be detected, thereby obtaining a virtual object reconstruction model; record the process of executing motion commands on the virtual object reconstruction model to obtain a target image corresponding to the content of the image to be detected; and perform clipping detection on the target image to obtain a new clipping detection result.

[0029] This application provides an electronic device for information detection, the electronic device comprising:

[0030] Memory is used to store executable instructions or computer programs.

[0031] The processor, when executing computer-executable instructions or computer programs stored in the memory, implements the information detection method provided in the embodiments of this application.

[0032] This application provides a computer-readable storage medium storing computer-executable instructions or computer programs, which, when executed by a processor, implement the information detection method provided in this application.

[0033] This application provides a computer program product, including computer-executable instructions or a computer program, which, when executed by a processor, implements the information detection method provided in this application.

[0034] The embodiments of this application have at least the following beneficial effects: for the image to be detected of the virtual object model, the clipping detection is performed by combining attention processing and image semantic information, thereby realizing a technical solution for automated clipping detection of the image to be detected of the virtual object model from multiple dimensions; thus, the accuracy and efficiency of clipping detection can be improved. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the architecture of the information detection system provided in the embodiments of this application;

[0036] Figure 2 This is one of the embodiments provided in this application. Figure 1 A schematic diagram of the terminal structure in the diagram;

[0037] Figure 3 This is a flowchart illustrating the information detection method provided in the embodiments of this application. Figure 1 ;

[0038] Figure 4This is a flowchart illustrating the information detection method provided in the embodiments of this application. Figure 2 ;

[0039] Figure 5 This is a flowchart illustrating the information detection method provided in the embodiments of this application. Figure 3 ;

[0040] Figure 6 This is a flowchart illustrating the information detection method provided in the embodiments of this application. Figure 4 ;

[0041] Figure 7 This is a flowchart illustrating the information detection method provided in the embodiments of this application. Figure 5 ;

[0042] Figure 8 This is a schematic diagram of a model training process provided in an embodiment of this application;

[0043] Figure 9 This is a schematic diagram of another model training process provided in an embodiment of this application;

[0044] Figure 10 This is a flowchart illustrating the information detection method provided in the embodiments of this application. Figure 6 ;

[0045] Figure 11 This is a schematic diagram of an exemplary process for detecting penetration provided in an embodiment of this application;

[0046] Figure 12 This is a schematic diagram of an exemplary penetration detection result provided in an embodiment of this application. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0048] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0049] In the following description, the terms “first, second, third, fourth” are used to distinguish similar objects and do not represent a specific ordering of objects. It is understood that “first, second, third, fourth” may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0050] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0051] Unless otherwise defined, all technical and scientific terms used in the embodiments of this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in the embodiments of this application is for the purpose of describing the embodiments of this application only and is not intended to limit this application.

[0052] In the implementation of this application, the collection and processing of relevant data should strictly comply with the requirements of relevant national laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0053] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0054] 1) Artificial neural networks are mathematical models that mimic the structure and function of biological neural networks. Exemplary structures of artificial neural networks in this application include Graph Convolutional Networks (GCNs, a type of neural network used to process data with graph structures), Deep Neural Networks (DNNs), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Neural State Machines (NSMs), and Phase-Functioned Neural Networks (PFNNs). In this application embodiment, the first training model, the second training model, the first clipping detection model, and the second clipping detection model all employ artificial neural networks.

[0055] 2) Clipping detection is a technique in 3D computer graphics. Clipping detection improves the rendering effect of virtual object models from multiple aspects, including visual effects, physical interaction, and collision detection. Among them, visual effects refer to the ability to detect the penetration phenomenon of rigid objects through clipping detection, thereby improving the realism and lifelikeness of the rendering; physical interaction refers to the ability to correctly handle the interaction between virtual object models through clipping detection; and collision detection refers to the ability to correctly handle the response and characteristics of virtual object models after collision through clipping detection.

[0056] 3) A virtual object model refers to the mathematical and computer graphics representation used to describe virtual objects in a virtual scene. A virtual object model includes not only the geometry and appearance of the virtual object, but also its physical properties, behavior, and interaction methods. Virtual object models can be applied in fields such as computer graphics, virtual reality (VR), augmented reality (AR), and computer-aided design (CAD).

[0057] 4) A virtual scene can be a simulation of the real world, a semi-simulated / semi-fictional virtual environment, or a purely fictional virtual environment. A virtual scene can be any of a two-dimensional, 2.5-dimensional, or three-dimensional virtual scene; this application does not limit the dimension of the virtual scene. For example, a virtual scene may include a virtual sky, virtual land, virtual ocean, etc. The virtual land may include environmental elements such as virtual deserts and virtual cities, and virtual objects can be controlled to move within the virtual scene.

[0058] 5) Virtual objects: These are interactive representations of people and objects within a virtual scene, or movable objects within the virtual scene. These movable objects can be virtual characters, virtual animals, anime characters, and virtual props, such as people and animals displayed in the virtual scene. A virtual object can also be a virtual avatar representing the user within the virtual scene. A virtual scene can include multiple virtual objects, each with its own shape and volume, occupying a portion of the space within the virtual scene.

[0059] It should be noted that, in order to perform clipping detection, manual review is usually adopted; that is, it is determined by humans whether the virtual object model in the virtual scene has clipping phenomenon; thus, the following technical problems exist.

[0060] 1. Impact on efficiency: Manual inspection requires time to check each model individually. In a batch of game resources, manual clipping detection may take weeks or even months.

[0061] 2. Affects accuracy: Due to the limited attention span of humans, even experienced developers may miss clipping in virtual object models.

[0062] 3. Limited scope of application: With the development of game technology, 3D game scenes are becoming more and more complex, and manual inspection is difficult to cope with this complexity and dynamic changes.

[0063] 4. Impact on flexibility: Manual testing is difficult to adapt quickly to game updates or the addition of new resources, which may lead to delays in the release of game updates.

[0064] In summary, the use of manual review affects the accuracy and efficiency of mold penetration detection.

[0065] In addition, a semi-automated method can be used for mold penetration detection; that is, the mold penetration detection tool is manually driven to perform the mold penetration detection. However, the mold penetration detection process still requires human intervention, which still affects the accuracy and efficiency of the mold penetration detection.

[0066] Based on this, embodiments of this application provide an information detection method, apparatus, electronic device, computer-readable storage medium, and computer program product, which can improve the accuracy and efficiency of through-molding detection. The following describes exemplary applications of the electronic device (referred to as the information detection device) for information detection provided in embodiments of this application. The information detection device provided in embodiments of this application can be implemented as various types of terminals such as robots, smartphones, smartwatches, laptops, tablets, desktop computers, smart home appliances, set-top boxes, smart vehicle devices, portable music players, personal digital assistants, dedicated messaging devices, intelligent voice interaction devices, portable gaming devices, and smart speakers. It can also be implemented as a server, or a combination of both; embodiments of this application do not limit this. The following will describe exemplary applications when the information detection device is implemented as a terminal.

[0067] See Figure 1 , Figure 1 This is a schematic diagram of the architecture of the information detection system provided in the embodiments of this application; as shown Figure 1 As shown, to support an information detection application, in the information detection system 100, a terminal 400 (terminals 400-1 and 400-2 are shown as examples) connects to a server 200 via a network 300. The network 300 can be a wide area network (WAN), a local area network (LAN), or a combination of both. The server 200 provides clipping detection services to the terminal 400 via the network 300 based on a trained first clipping detection model and a second clipping detection model. Additionally, the information detection system 100 also includes a database 500 for providing data support to the server 200; and... Figure 1 The example shown illustrates a scenario where the database 500 is independent of the server 200. However, the database 500 can also be integrated into the server 200, and this embodiment does not limit this to any particular case.

[0068] Terminal 400 is used to perform attention processing on the initial image features of the image to be detected to obtain the image features to be detected, wherein the image to be detected is an image of a virtual object model; to perform clipping detection on the image features to be detected to obtain a first clipping detection result; to construct an element graph network based on the image semantic information of the image to be detected, wherein the element graph network represents the semantic relationship between each image element in the image to be detected; to perform clipping detection on the element graph network to obtain a second clipping detection result; and to combine the first clipping detection result and the second clipping detection result to obtain the target clipping detection result of the image to be detected, and to display the target clipping detection result (exemplary graphical interfaces 400-11 and 400-21 for displaying the target clipping detection result are shown).

[0069] In some embodiments, server 200 may be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. Terminals and servers can be connected directly or indirectly via wired or wireless communication, which is not limited in this embodiment.

[0070] See Figure 2 , Figure 2 This is one of the embodiments provided in this application. Figure 1 A schematic diagram of the terminal structure in the diagram; such as Figure 2 As shown, terminal 400 includes at least one processor 410, memory 450, at least one network interface 420, and user interface 430. The various components in terminal 400 are coupled together via a bus system 440. It is understood that the bus system 440 is used to implement communication between these components. In addition to a data bus, the bus system 440 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 2 The general labeled all buses as Bus System 440.

[0071] The processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0072] User interface 430 includes one or more output devices 431 that enable the presentation of media content, including one or more speakers and / or one or more visual displays. User interface 430 also includes one or more input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0073] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state storage, hard disk drives, optical disk drives, etc. The memory 450 may optionally include one or more storage devices physically located away from the processor 410.

[0074] The memory 450 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), and the volatile memory may be random access memory (RAM). The memory 450 described in this application embodiment is intended to include any suitable type of memory.

[0075] In some embodiments, memory 450 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0076] Operating system 451 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, driver layer, etc., for implementing various basic business functions and handling hardware-based tasks;

[0077] The network communication module 452 is used to reach other electronic devices via one or more (wired or wireless) network interfaces 420, exemplary network interfaces 420 including: Bluetooth, Wi-Fi, and Universal Serial Bus (USB), etc.

[0078] Presentation module 453 is configured to enable the presentation of information (e.g., a user interface for operating peripheral devices and displaying content and information) via one or more output devices 431 (e.g., a display screen, a speaker, etc.) associated with user interface 430;

[0079] The input processing module 454 is used to detect and translate one or more user inputs or interactions from one or more input devices 432.

[0080] In some embodiments, the information detection device provided in this application can be implemented in software. Figure 2 An information detection device 455 stored in memory 450 is shown. This device can be software in the form of programs and plug-ins, and includes the following software modules: feature processing module 4551, first detection module 4552, network construction module 4553, second detection module 4554, result acquisition module 4555, model training module 4556, and model reconstruction module 4557. These modules are logically connected and can therefore be arbitrarily combined or further split according to their implemented functions. The functions of each module will be described below.

[0081] In some embodiments, the information detection device provided in this application can be implemented in hardware. As an example, the information detection device provided in this application can be a processor in the form of a hardware decoding processor, which is programmed to execute the information detection method provided in this application. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0082] In some embodiments, the terminal or server can implement the information detection method provided in this application by running various computer-executable instructions or computer programs. For example, computer-executable instructions can be microprogram-level commands, machine instructions, or software instructions. Computer programs can be native programs or software modules in an operating system; they can be native applications (APPs), i.e., programs that need to be installed in the operating system to run, such as game testing APPs or animation production APPs; or they can be applets that can be embedded in any APP, i.e., programs that only need to be downloaded to a browser environment to run. In summary, the aforementioned computer-executable instructions can be any form of instruction, and the aforementioned computer programs can be any form of application, module, or plugin.

[0083] The information detection method provided in this application will be described below with reference to exemplary applications and implementations of the information detection equipment provided in the embodiments of this application. Furthermore, the information detection method provided in the embodiments of this application is applicable to various penetration detection scenarios such as cloud technology, artificial intelligence, smart transportation, and games.

[0084] See Figure 3 , Figure 3 This is a flowchart illustrating the information detection method provided in the embodiments of this application. Figure 1 ,in, Figure 3 The main body responsible for executing each step is the information detection equipment; the following will combine... Figure 3 The steps shown are explained.

[0085] Step 101: Perform attention processing on the initial image features of the image to be detected to obtain the image features to be detected.

[0086] In this embodiment, once the virtual object model is constructed, clipping detection begins. Here, the information detection device performs clipping detection on the virtual object model using its image, which is referred to as the image to be detected. Furthermore, the information detection device performs clipping detection on the image to be detected from multiple dimensions, one of which is the feature attention dimension. When performing clipping detection on the image to be detected based on the feature attention dimension, the information detection device first extracts the initial image features of the image to be detected, and then performs attention processing on the initial image features to enhance the features that are helpful for clipping detection. The attention processing result of the obtained initial image features is the feature of the image to be detected.

[0087] It should be noted that the image to be detected is an image of a virtual object model. This can be a single image of the virtual object model, or any image during the process of controlling the virtual object model to perform a specified action, etc. This application embodiment does not limit this. The virtual object model can be, for example, a virtual racing car model, a game character model, an animated character model, etc. The initial image features are features extracted from the image to be detected for clipping detection. These features can be extracted using a trained neural network model, extracted using a specified encoding method, extracted using a large language model, or a combination of the above, etc. This application embodiment does not limit this. Attention processing, also known as an attention mechanism, refers to a way of simulating human attention concentration. It can assign different weights to different parts of the initial image features based on their importance, thereby focusing more on the features used for clipping detection compared to the initial image features. In other words, attention processing is used to assign different weights to different parts of the initial image features in order to process the initial image features in a targeted manner. Here, attention processing can be self-attention, multi-head attention, or a combination of both, etc., and the embodiments of this application do not limit it in this way.

[0088] See Figure 4 , Figure 4 This is a flowchart illustrating the information detection method provided in the embodiments of this application. Figure 2 ,in, Figure 4 The main entity executing each step is the information detection equipment; such as... Figure 4 As shown in the embodiment of this application, step 101 can be implemented by steps 1011 and 1012; that is, the information detection device performs attention processing on the initial image features of the image to be detected to obtain the image features to be detected, including steps 1011 and 1012. Each step will be described below.

[0089] Step 1011: Perform attention processing on the initial image features of the image to be detected to obtain the attention weights of the initial image features.

[0090] It should be noted that when the information detection device performs attention processing on the initial image features, it first assigns weights to the initial image features. The resulting weights are the attention weights of the initial image features. It is easy to see that the attention weights represent the importance of different parts of the initial image features in the clipping detection process, and are used to enhance features in the initial image that are helpful for clipping detection. Here, the sum of the attention weights can be a specified value, such as 1.

[0091] In this embodiment of the application, the attention weight is obtained by the information detection device based on the attention mechanism to assign weights to the initial image features. The process of weight assignment based on the attention mechanism includes first defining the query, key, and value, then calculating the attention score based on the query, key, and value. The attention score represents the association strength between the query and each key. Next, scaling is performed based on the attention score. Finally, the scaling result is normalized to obtain the attention weight and complete the weight assignment.

[0092] Step 1012: Fuse the initial image features and attention weights to obtain the features of the image to be detected.

[0093] It should be noted that the information detection device fuses the initial image features and attention weights, and the fused result is the image feature to be detected. Here, the information detection device can use a corresponding multiplication method to fuse the initial image features and attention weights.

[0094] In the embodiments of this application, attention processing can be performed once or iteratively, and this application does not limit the specific implementation. When attention processing is performed iteratively, after obtaining the fusion result of the initial image features and attention weights, the information detection device performs attention processing on the fusion result of the initial image features and attention weights until the condition for the end of attention processing iteration is met, and then determines the final fusion result as the image feature to be detected. The end condition of attention processing iteration can be a specified iteration duration, a specified number of iterations, or a combination of the above, etc., and this application does not limit the specific implementation.

[0095] It is understandable that by performing attention processing on the initial image features, dynamic weight allocation is achieved for different parts of the initial image features, which enhances the importance of features related to clipping detection and reduces the importance of features unrelated to clipping detection, thereby increasing the importance of features that are helpful for clipping detection in the initial image features.

[0096] Step 102: Perform clipping detection on the features of the image to be detected to obtain the first clipping detection result.

[0097] In this embodiment, after the information detection device obtains the features of the image to be detected, it also obtains features for clipping detection. Therefore, the information detection device performs clipping detection based on the features of the image to be detected, and the obtained clipping detection result is the first clipping detection result. Here, the information detection device can use a first activation function (e.g., the sigmoid function, the softmax function) to perform clipping detection on the features of the image to be detected to obtain the first clipping detection result; wherein, the first activation function is a trained activation function used for clipping detection based on the features of the image to be detected.

[0098] It should be noted that the first clipping detection result represents the clipping detection result corresponding to the features of the image to be detected. Since the features of the image to be detected are obtained by performing attention processing on the initial image features of the image to be detected, the first clipping detection result represents the clipping detection result of the image to be detected based on the attention processing. The first clipping detection result can indicate that the virtual object model in the image to be detected has clipping (also known as clipping phenomenon), or it can indicate that the virtual object model in the image to be detected does not have clipping.

[0099] It should also be noted that the first clipping detection result is the clipping detection result of the image to be detected obtained from the attention processing dimension. The information detection device combines the first clipping detection result with the clipping detection results of the image to be detected obtained from other dimensions to jointly determine the final clipping detection result of the image to be detected.

[0100] It is understandable that in the process of obtaining the first clipping detection result of the image to be detected based on attention processing, attention processing can accurately extract the effective features from the initial image features of the image to be detected, thereby improving the accuracy and efficiency of clipping detection.

[0101] Step 103: Construct an element graph network based on the image semantic information of the image to be detected.

[0102] In this embodiment of the application, the information detection device can perform clipping detection on the image to be detected from the semantic information dimension of the image to be detected. When the information detection device performs clipping detection on the image to be detected from the semantic information dimension, it first obtains the semantic information of the image to be detected and calls the obtained semantic information of the image to be detected "image semantic information". Then, based on each image element in the image semantic information and the semantic relationship between each image element, a graph network is constructed. The constructed graph network is called an element graph network.

[0103] It should be noted that image semantic information represents the text description of the image to be detected. The element graph network represents the semantic relationships between the various image elements in the image to be detected. The semantic relationships include at least one of the following: hierarchical relationship, positional relationship, and action relationship; the hierarchical relationship represents the display hierarchy between the various image elements, the positional relationship represents the relative position between the various image elements, and the action relationship represents the rigid body motion between the various image elements.

[0104] See also Figure 4 In this embodiment of the application, step 103 includes steps 1031 and 1032; that is, the information detection device constructs an element graph network based on the image semantic information of the image to be detected, including steps 1031 and 1032. Each step is described below.

[0105] Step 1031: Determine each image element of the image to be detected and the semantic relationships between each image element from the image semantic information.

[0106] It should be noted that since image semantic information is a textual description of the image to be detected, it includes each image element in the image. For example, when the image to be detected is a virtual race car crashing into a virtual wall, the virtual race car and the virtual wall are the image elements in the image. Therefore, the information detection device can determine each image element of the image to be detected from the image semantic information. Furthermore, since image semantic information is a textual description of the image to be detected, it also includes the semantic relationships between the image elements. For example, when the image to be detected is a virtual race car crashing into a virtual wall, the crash is the semantic relationship between the image elements. Therefore, the information detection device can determine the semantic relationships between the image elements of the image to be detected from the image semantic information.

[0107] Step 1032: Using each image element as a graph network node and constructing graph network node edges based on semantic relationships, an element graph network is obtained.

[0108] In this embodiment of the application, the information detection device uses each image element in each image element as a graph network node, and constructs an edge (also known as a graph network node edge) between two graph network nodes based on semantic relationships; after all graph network nodes and graph network node edges are determined, an element graph network is constructed.

[0109] It is understandable that by constructing an element graph network using the image semantic information of the image to be detected, the process of representing the semantic relationships between various image elements in the image to be detected through a graph network is realized, thereby improving the accuracy of image content parsing of the image to be detected.

[0110] Step 104: Perform clipping detection on the element graph network to obtain the second clipping detection result.

[0111] In this embodiment, the information detection device first obtains corresponding features based on the element-wise graph network, thus obtaining the graph network features. Then, the information detection device performs clipping detection based on the graph network features, and the obtained clipping detection result is the second clipping detection result. Here, the information detection device can use a second activation function (e.g., the sigmoid function, the Sofimax function) to perform clipping detection on the features of the image to be detected, to obtain the second clipping detection result; wherein, the second activation function is a pre-trained activation function used for clipping detection based on graph network features.

[0112] It should be noted that the second clipping detection result represents the clipping detection result corresponding to the element graph network. Since the element graph network is obtained by constructing a graph network based on the image semantic information of the image to be detected, the second clipping detection result represents the clipping detection result of the image to be detected based on the image semantic information. The second clipping detection result can indicate that the virtual object model in the image to be detected has clipping, or it can indicate that the virtual object model in the image to be detected does not have clipping.

[0113] It should also be noted that the second clipping detection result is the clipping detection result of the image to be detected obtained from the semantic information dimension of the image. The information detection device will combine the second clipping detection result with the clipping detection results of the image to be detected obtained from other dimensions to jointly determine the final clipping detection result of the image to be detected.

[0114] It is understandable that in the process of obtaining the second clipping detection result of the image to be detected based on image semantic information, the image semantic information can accurately describe the image content of the image to be detected, thereby improving the accuracy and efficiency of clipping detection.

[0115] In step 104 of this application embodiment, the information detection device performs clipping detection on the element graph network to obtain a second clipping detection result, including: the information detection device first performs clipping detection on the element graph network to obtain a third clipping detection result; then obtains N fourth clipping detection results corresponding to the first N video frame images of the image to be detected; finally, the third clipping detection result and the N fourth clipping detection results are combined to obtain the second clipping detection result.

[0116] It should be noted that the third clipping detection result refers to the clipping detection result obtained by performing clipping detection on the element graph network. When the image to be detected is any frame of video in a video of a virtual object model executing a specified motion command, the information detection device acquires the clipping detection results corresponding to the historical frame video images of the image to be detected, which are the N fourth clipping detection results corresponding to the previous N video frame images of the image to be detected. The N video frame images correspond one-to-one with the N fourth clipping detection results. The second clipping detection result is determined by combining the third clipping detection result and the N fourth clipping detection results. Here, N is a positive integer, such as 1, 3, etc.

[0117] In this embodiment, the information detection device can determine that the second clipping detection result indicates that the virtual object model in the image to be detected has clipping when both the third clipping detection result and N fourth clipping detection results indicate that the virtual object model has clipping; and determine that the second clipping detection result indicates that the virtual object model in the image to be detected does not have clipping when at least one of the third clipping detection result and N fourth clipping detection results indicates that the virtual object model in the image to be detected does not have clipping.

[0118] In the embodiments of this application, the execution order of steps 101 and 102 is independent of the execution order of steps 103 and 104; therefore, the execution of steps 101 and 102 can be synchronous, sequential, or alternating with the execution of steps 103 and 104, etc., and the embodiments of this application do not limit this.

[0119] Understandably, by combining the third clipping detection result and N fourth clipping detection results to determine the second clipping detection result, the historical clipping detection results obtained based on semantic information and the current clipping detection results are integrated, thus improving the accuracy of the second clipping detection result.

[0120] Step 105: Combine the first and second clipping detection results to obtain the target clipping detection result of the image to be detected.

[0121] In this embodiment of the application, the information detection device determines the final clipping detection result of the image to be detected from two dimensions: attention processing dimension and image semantic information dimension. Thus, after obtaining the first clipping detection result corresponding to the attention processing dimension and the second clipping detection result corresponding to the image semantic information dimension, the information detection device combines the first clipping detection result and the second clipping detection result to jointly determine the final clipping detection result of the image to be detected, and refers to the final clipping detection result of the image to be detected as the target clipping detection result.

[0122] It should be noted that the target clipping detection result represents the final clipping detection result of the image to be detected; here, the target clipping detection result can indicate that the virtual object model in the image to be detected has clipping, or it can indicate that the virtual object model in the image to be detected does not have clipping, or it can indicate whether the virtual object model in the image to be detected has clipping to be confirmed.

[0123] See also Figure 4 In this embodiment of the application, step 105 includes step 1051A, step 1051B, or step 1051C; that is, the information detection device combines the first clipping detection result and the second clipping detection result to obtain the target clipping detection result of the image to be detected, including step 1051A, step 1051B, or step 1051C. Each step will be described below.

[0124] Step 1051A: When both the first and second clipping detection results indicate that the virtual object model is clipping through the image, the information that the virtual object model is clipping through the image to be detected is determined as the target clipping detection result.

[0125] It should be noted that when the first clipping detection result indicates that the virtual object model in the image to be detected has clipping, and the second clipping detection result also indicates that the virtual object model in the image to be detected has clipping, that is, when both the first and second clipping detection results indicate that the virtual object model has clipping, the information detection device determines that the virtual object model in the image to be detected has clipping, and thus determines the information that the virtual object model in the image to be detected has clipping as the target clipping detection result.

[0126] Step 1051B: When both the first and second clipping detection results indicate that the virtual object model is not clipped, the information that there is no clipping of the virtual object model in the image to be detected is determined as the target clipping detection result.

[0127] It should be noted that when the first clipping detection result indicates that the virtual object model in the image to be detected does not clip, and the second clipping detection result also indicates that the virtual object model in the image to be detected does not clip, that is, when both the first and second clipping detection results indicate that the virtual object model does not clip, the information detection device determines that there is no virtual object model clipping in the image to be detected, and thus determines the information that there is no virtual object model clipping in the image to be detected as the target clipping detection result.

[0128] Step 1051C: When one of the first and second clipping detection results indicates that the virtual object model is clipping, the information of the virtual object model to be clipped in the image to be detected is determined as the target clipping detection result.

[0129] It should be noted that when the information detection device determines whether the virtual object model in the image under test has clipping due to clipping, and the second clipping detection result indicates that the virtual object model in the image under test does not have clipping due to clipping, or when the first clipping detection result indicates that the virtual object model in the image under test does not have clipping due to ...

[0130] In this embodiment of the application, when the target clipping detection result indicates that the virtual object model in the image to be detected needs to be verified by clipping detection, the image to be detected can be determined as an image to be manually reviewed or to be re-detected by clipping; and then the image to be detected can be manually reviewed or re-detected by clipping.

[0131] Understandably, by combining attention processing and image semantic information to perform clipping detection on images of virtual object models, a technical solution for automated clipping detection of images of virtual object models from multiple dimensions is achieved; thus, the accuracy and efficiency of clipping detection can be improved.

[0132] See Figure 5 , Figure 5 This is a flowchart illustrating the information detection method provided in the embodiments of this application. Figure 3 ,in, Figure 5 The main entity executing each step is the information detection equipment; such as... Figure 5As shown in the embodiment of this application, steps 106 and 107 are included before step 103; that is, before the information detection device constructs the element graph network based on the image semantic information of the image to be detected, the information detection method further includes steps 106 and 107. Each step will be described below.

[0133] Step 106: Extract the semantic features of the image to be detected.

[0134] It should be noted that the information detection device can obtain the semantic information of the image to be detected by generating text from the image. That is, the information detection device first identifies information such as the subject, scene, and activities in the image to be detected, and then extracts features from the identified information, thus obtaining the image semantic features. Here, the image semantic features represent the features of the image to be detected in the semantic information dimension. The information detection device can use a deep learning model (e.g., CNN) to extract the image semantic features.

[0135] Step 107: Determine the image semantic information of the image to be detected based on image semantic features.

[0136] It should be noted that the information detection device converts the extracted image semantic features into natural language descriptions, and the resulting conversion is the image semantic information of the image to be detected. Therefore, the method of generating text from images includes image information recognition, feature extraction, and natural language description conversion. Here, the information detection device can use natural language processing (NLP) to convert image semantic features into natural language descriptions, such as using recurrent neural networks, long short-term memory networks (LSTM), or generative adversarial networks (GANs).

[0137] It is understandable that by extracting the semantic features of the image to be detected in the semantic information dimension, and then converting the semantic features into natural language descriptions to obtain the semantic information of the image, the accuracy of obtaining the semantic information of the image is improved.

[0138] See Figure 6 , Figure 6 This is a flowchart illustrating the information detection method provided in the embodiments of this application. Figure 4 ,in, Figure 6 The main entity executing each step is the information detection equipment; such as... Figure 6 As shown in the embodiment of this application, steps 108 to 110 are included before step 101; that is, before the information detection device performs attention processing on the initial image features of the image to be detected to obtain the image features to be detected, the information detection method further includes steps 108 to 110. Each step is described below.

[0139] Step 108: Record the process of executing motion commands on the virtual object model to obtain the video to be tested.

[0140] It should be noted that when the information detection equipment receives a motion command for the virtual object model, it controls the virtual object model to execute the motion command; and it records the process of the virtual object model executing the motion command. The resulting video recording is the video to be detected; thus, the video to be detected represents the process of the virtual object model executing the motion command. Here, the motion command refers to the instruction that controls the movement of the virtual object model, such as the instruction to move along a specified virtual road.

[0141] Step 109: Divide the video to be detected into video frames to obtain a sequence of video frame images.

[0142] It should be noted that after obtaining the video to be detected, the information detection device divides the video into video frames to obtain a sequence of video frame images corresponding to the video to be detected. This video frame division can be performed by extracting each frame of the video to be detected, extracting keyframes, extracting video frames based on a specified video frame interval, or a combination of the above, etc. This application embodiment does not limit the specific method used.

[0143] Step 110: Traverse the video frame image sequence and identify the traversed video frame images as the images to be detected.

[0144] It should be noted that after obtaining the video frame image sequence, the information detection device traverses the video frame image sequence and identifies each traversed video frame image as the image to be detected, in order to perform clipping detection on the image to be detected. Here, the video frame image refers to the sequence element in the video frame image sequence.

[0145] It is understandable that by acquiring the video of the virtual object model executing motion commands, and then performing clipping detection on any image in the video, clipping detection can be achieved at any instant when the virtual object model executes motion commands, thereby improving the precision and accuracy of clipping detection.

[0146] See Figure 7 , Figure 7 This is a flowchart illustrating the information detection method provided in the embodiments of this application. Figure 5 ,in, Figure 7 The main entity executing each step is the information detection equipment; such as... Figure 7As shown in the embodiment of this application, steps 111 and 112 are included before step 101; that is, before the information detection device performs attention processing on the initial image features of the image to be detected to obtain the image features to be detected, the information detection method further includes steps 111 and 112. Each step will be described below.

[0147] Step 111: Perform image preprocessing on the image to be detected to obtain the image to be processed.

[0148] It should be noted that the image to be processed is the result of image preprocessing of the image to be detected. Image preprocessing includes at least one of the following: noise reduction, brightness adjustment, and contrast adjustment.

[0149] It should also be noted that denoising, brightness adjustment, and contrast adjustment are used to improve image quality and enhance visual effects. Denoising refers to the process of reducing noise in the image to be inspected. Noise can be caused by various reasons, such as camera sensor accuracy, transmission errors, or environmental factors. Denoising improves the clarity of the image. Information inspection equipment can use at least one of mean filtering, median filtering, and bilateral filtering for denoising. Mean filtering reduces noise by averaging the values ​​of the pixels surrounding each pixel in the image to be inspected. Median filtering replaces the pixel's value by selecting the median value of the pixels surrounding the pixel in the image to be inspected. Bilateral filtering determines the pixel value by combining spatial proximity and pixel value similarity in the image to be inspected. Brightness adjustment enhances visual effects by changing the overall brightness of the image. Information inspection equipment can use at least one of linear transformation and nonlinear transformation for brightness adjustment. Linear transformation adjusts the brightness by adding or subtracting a constant to each pixel in the image to be inspected, while nonlinear transformation adjusts the brightness of the image to be inspected nonlinearly. Contrast adjustment refers to the process of changing the distribution range of pixel values ​​in the image to be detected, which can enhance the difference between different regions in the image to be detected and improve the visual effect of the image to be detected. The information detection device can use at least one of linear contrast enhancement, histogram equalization and sharpening to perform contrast adjustment. Linear contrast enhancement improves contrast by stretching or compressing the distribution range of pixel values ​​in the image to be detected. Histogram equalization equalizes the histogram of the image to be detected so that the probability of each gray level appearing in the image to be detected is approximately equal. Sharpening is used to enhance the edges and textures in the image to be detected and improve the clarity of the image to be detected.

[0150] Step 112: Extract features from the image to be processed to obtain initial image features.

[0151] It should be noted that the information detection equipment extracts features from the image to be processed in the dimension of clipping detection, and the obtained feature extraction result is the initial image feature. Therefore, the initial image feature represents the feature of the image to be processed in the dimension of clipping detection.

[0152] It is understandable that by performing image preprocessing on the image to be detected before extracting features for the clipping detection dimension, the accuracy of the initial image features can be improved.

[0153] In this embodiment, the first clipping detection result can be achieved through a first clipping detection model, which is used to perform clipping detection on the image to be detected from the feature attention dimension; see also Figure 8 , Figure 8 This is a schematic diagram of a model training process provided in an embodiment of this application, wherein, Figure 8 The main entity executing each step is the information detection equipment; such as... Figure 8 As shown in the embodiment of this application, the first penetration detection model is obtained through training in steps 113 to 115. Each step is described below.

[0154] Step 113: Obtain the first image sample and the first clipping annotation result of the first image sample.

[0155] It should be noted that the information detection device acquires first training data, which includes first image samples and first clipping annotation results; wherein, the first image sample refers to the image sample to be detected for clipping, which may be an image with clipping or an image without clipping, and this application embodiment does not limit this; the first clipping annotation result is the result of whether clipping exists for the first image sample, which is a real result of whether clipping exists, and may be either present or absent.

[0156] Step 114: Use the first model to be trained to perform clipping detection on the first image sample to obtain the first clipping prediction result.

[0157] In this embodiment, the information detection device uses a first training model to perform clipping detection on the first image sample, and the obtained clipping detection result is the first clipping prediction result. Here, the process of the information detection device using the first training model to perform clipping detection on the first image sample is similar to the process described in steps 101 and 102 of the information detection device obtaining the first clipping detection result of the image to be detected, and will not be repeated here.

[0158] It should be noted that the first model to be trained is a neural network model to be trained for clipping detection based on attention processing. It can be the original neural network model, a pre-trained neural network model, or a large model, etc., and this application does not limit it in this respect. The first clipping prediction result represents the clipping detection result of the first image sample obtained by using the first model to be trained.

[0159] Step 115: Based on the difference between the first clipping prediction result and the first clipping annotation result, train the first training model to obtain the first clipping detection model.

[0160] In this embodiment, since the first model to be trained is a neural network model to be trained, the accuracy of the first clipping prediction result obtained using the first model to be trained needs to be compared and confirmed with the first clipping annotation result. Therefore, the information detection device compares the first clipping prediction result with the first clipping annotation result, and calculates a first loss function value based on the difference between the two results. Then, backpropagation is performed on the first model to be trained based on this first loss function value to adjust the model parameters and train the model. Finally, the trained first model to be trained is called the first clipping detection model. Backpropagation is a technique used to train neural network models. In a supervised learning environment, backpropagation calculates the gradient of the loss function value with respect to the model parameters and updates the model parameters through gradient descent to minimize the loss function value.

[0161] It should be noted that the training process of the first model to be trained can be a single iteration or an iterative process, and this embodiment does not limit this. When the training process of the first model to be trained is iterative, the iterative training ends when the first iteration termination condition is met, and the first model to be trained in the last iteration is determined as the first clipping detection model. The first iteration termination condition can be reaching a first accuracy threshold, a first iteration count threshold, a first iteration duration threshold, or a combination of the above, etc., and this embodiment does not limit this.

[0162] In this embodiment, the calculation of the first loss function value can be achieved through a first loss function; the first loss function refers to the loss function that calculates the loss function value between the first pattern-breaking prediction result and the first pattern-breaking annotation result.

[0163] It should be noted that loss functions in machine learning and deep learning are used to quantify the difference between model predictions and actual values; loss functions include mean squared error (MSE), mean absolute error (MAE), cross-entropy loss, and log loss.

[0164] In this embodiment, the second clipping detection result can be achieved through a second clipping detection model, which is used to perform clipping detection on the image to be detected from the dimension of image semantic information; see also Figure 9 , Figure 9 This is a schematic diagram of another model training process provided in an embodiment of this application, wherein, Figure 9 The main entity executing each step is the information detection equipment; such as... Figure 9 As shown in the embodiment of this application, the second penetration detection model is obtained through training in steps 116 to 118. Each step is described below.

[0165] Step 116: Obtain the second image sample and the second clipping annotation result of the second image sample.

[0166] It should be noted that the information detection device acquires second training data, which includes second image samples and second clipping annotation results; wherein, the second image sample refers to the image sample to be detected for clipping, which may be an image with clipping or an image without clipping, and this application embodiment does not limit this; the second clipping annotation result is the result of whether clipping exists in the second image sample, which is a real result of whether clipping exists, and may be either present or absent.

[0167] It should also be noted that the second image sample may be the same as or different from the first image sample, and this application embodiment does not limit this; when the second image sample is the same as the first image sample, the first clipping annotation result is also the same as the second clipping annotation result.

[0168] Step 117: Use the second model to be trained to perform clipping detection on the second image sample to obtain the second clipping prediction result.

[0169] In this embodiment, the information detection device uses a second training model to perform clipping detection on the second image sample, and the obtained clipping detection result is the second clipping prediction result. Here, the process of the information detection device using the second training model to perform clipping detection on the second image sample is similar to the process described in steps 103 and 104 of the information detection device obtaining the second clipping detection result of the image to be detected, and will not be repeated here.

[0170] It should be noted that the second model to be trained is a neural network model to be trained for clipping detection based on semantic information. It can be the original neural network model, a pre-trained neural network model, or a large model, etc., and this application does not limit it in this respect. The second clipping prediction result represents the clipping detection result of the second image sample obtained by using the second model to be trained.

[0171] Step 118: Based on the difference between the second clipping prediction result and the second clipping annotation result, train the second training model to obtain the second clipping detection model.

[0172] In this embodiment, since the second model to be trained is a neural network model to be trained, the accuracy of the second clipping prediction result obtained using the second model to be trained needs to be compared and confirmed with the second clipping annotation result. Therefore, the information detection device compares the second clipping prediction result with the second clipping annotation result, and calculates the second loss function value based on the difference between the second clipping prediction result and the second clipping annotation result. Then, based on the second loss function value, backpropagation is performed in the second model to be trained to adjust the model parameters in the second model to be trained, thereby achieving the training of the second model to be trained. Finally, the trained second model to be trained is called the second clipping detection model.

[0173] It should be noted that the training process of the second model to be trained can be a single iteration or an iterative process, and this embodiment does not limit this. When the training process of the second model to be trained is iterative, the iterative training ends when the second iteration termination condition is met, and the second model to be trained in the last iteration is determined as the second clipping detection model. The second iteration termination condition can be reaching the second accuracy index threshold, the second iteration number threshold, the second iteration duration threshold, or a combination of the above, etc., and this embodiment does not limit this.

[0174] In this embodiment, the calculation of the second loss function value can be achieved through a second loss function; the second loss function refers to the loss function that calculates the loss function value between the second mold-breaking prediction result and the second mold-breaking annotation result.

[0175] See Figure 10 , Figure 10 This is a flowchart illustrating the information detection method provided in the embodiments of this application. Figure 6 ,in, Figure 10 The main entity executing each step is the information detection equipment; such as... Figure 10 As shown in the embodiment of this application, steps 119 to 121 are included after step 105; that is, before the information detection device performs attention processing on the initial image features of the image to be detected to obtain the image features to be detected, the information detection method further includes steps 119 to 121. Each step is described below.

[0176] Step 119: When the target clipping detection result indicates that there is a virtual object model clipping in the image to be detected, the virtual object model is reconstructed based on the image to be detected to obtain the virtual object reconstruction model.

[0177] It should be noted that when the target clipping detection result indicates that there is a virtual object model clipping in the image to be detected, it indicates that there is a problem with the virtual object model in the image to be detected. Therefore, the information detection device determines the reconstruction information of the virtual object model based on the information represented by the image to be detected, and then performs model reconstruction based on the reconstruction information. The obtained model reconstruction result is called the virtual object reconstruction model.

[0178] Step 120: Record the process of executing motion commands on the virtual object reconstruction model to obtain the target image corresponding to the content of the image to be detected.

[0179] It should be noted that the process of the information detection device recording video of the motion commands executed on the virtual object reconstruction model is similar to the process of recording video of the motion commands executed on the virtual object model, and will not be described again in this embodiment. Here, the information detection device uses the video frame image corresponding to the content of the image to be detected in the recorded video as the target image.

[0180] Step 121: Perform clipping detection on the target image to obtain new clipping detection results.

[0181] It should be noted that the information detection equipment can perform clipping detection on the target image manually to obtain a new clipping detection result, or it can perform clipping detection on the target image using the same method as performing clipping detection on the image to be detected, etc. This application embodiment does not limit this approach. Here, the information detection equipment refers to the clipping detection result of the target image as the new clipping detection result. Furthermore, when performing clipping detection on the target image using the same method as performing clipping detection on the image to be detected, since the process of performing clipping detection on the target image is similar to the process of performing clipping detection on the image to be detected, this application embodiment will not repeat the description here.

[0182] It is understandable that when the phenomenon of virtual object model clipping is determined in the image to be detected, the information to be reconstructed of the virtual object model is determined based on the information represented by the image to be detected. Then, the virtual object model is reconstructed based on the information to be reconstructed, and targeted model detection is performed on the reconstructed model, thereby improving the modeling quality of the virtual object.

[0183] The following describes an exemplary application of the embodiments of this application in a real-world application scenario. This exemplary application describes the process of clipping detection for game models in a game scene. It is readily apparent that the information detection method provided by the embodiments of this application can be applied to clipping detection in game scenes, and also to clipping detection in virtual scenes such as animation scenes and simulation scenes. Here, the application in a game scene is used as an example for illustration.

[0184] It should be noted that game scenarios refer to virtual environments used for game matches (referred to as virtual matches), such as virtual racing scenarios, scenarios in first-person shooter (FPS) games, scenarios in third-person shooter (TPS) games, and scenarios in multiplayer online battle arena (MOBA) games.

[0185] See Figure 11 , Figure 11 This is a schematic diagram of an exemplary penetration detection process provided in an embodiment of this application; as shown... Figure 11As shown, this exemplary clipping detection process includes five steps: clipping resource collection 11-1, clipping resource preprocessing 11-2, clipping detection based on attention mechanism of preprocessed results 11-3, clipping detection based on semantic information of preprocessed results 11-4, and clipping detection result acquisition 11-5. Clipping resource collection 11-1 refers to recording video of the 3D game model to be detected (referred to as the virtual object model), and the recorded video (referred to as the video to be detected) is the collected clipping resource. Clipping resource preprocessing 11-2 (referred to as image preprocessing) refers to the process of preprocessing the recorded video frame by frame, and the preprocessing result is the preprocessed video frame image corresponding to each frame. The first step is to perform clipping detection on the preprocessed video frame image (referred to as the image to be processed). Step 11-3 refers to using a neural network model including an attention mechanism to perform clipping detection on each preprocessed video frame image to obtain clipping detection result 11-61 (referred to as the first clipping detection result). Step 11-4 refers to performing clipping detection based on the semantic information of the preprocessed video frame image to obtain clipping detection result 11-62 (referred to as the second clipping detection result). Finally, step 11-5 refers to combining clipping detection result 11-61 and clipping detection result 11-62 to obtain the final clipping detection result 11-7 (referred to as the target clipping detection result). Each processing step is explained below.

[0186] First, let's explain how to collect clipping resources.

[0187] It should be noted that the 3D game model can realize specified virtual functions, such as driving along a virtual track or driving a virtual race car along a virtual track. When the 3D game model performs the specified virtual function, the process of the 3D game model performing the specified virtual function is recorded on video, which completes the collection of clipping resources.

[0188] Next, the preprocessing of the molding resources will be explained.

[0189] In the process of extracting each frame of a recorded video for preprocessing, the video frame images are first extracted from the recorded video, and then each video frame image is preprocessed to obtain the preprocessed video frame images. Preprocessing includes at least one of the following: noise reduction, brightness adjustment, and contrast enhancement.

[0190] In the embodiments of this application, the preprocessing can be described by formula (1), which is shown below.

[0191] I' t =P(I t (1);

[0192] Among them, I t Let I' represent the original t-th video frame image (called the image to be detected), P represent preprocessing, and I' represent... t This represents the preprocessed t-th video frame image (referred to as the image to be processed).

[0193] Next, the process of performing clipping detection on the preprocessed results based on the attention mechanism will be explained.

[0194] It should be noted that clipping detection based on the attention mechanism in the preprocessed results can be achieved using a neural network model that includes the attention mechanism (referred to as the first clipping detection network). The attention mechanism is a neural network model used to simulate human visual attention. Through the attention mechanism, different weights can be assigned to different parts of an image, thereby highlighting the importance of features in the image. Specifically, the attention mechanism learns to determine the importance of each part of the image.

[0195] In this embodiment, the process of performing clipping detection on the preprocessed results based on the attention mechanism includes feature extraction, weight allocation, feature determination, and clipping detection. Feature extraction refers to extracting features from the preprocessed video frame image; weight allocation refers to assigning weights to the extracted features; feature determination refers to determining the features to be detected based on the assigned weights; and clipping detection refers to performing clipping detection on the features to be detected to obtain the corresponding clipping detection result.

[0196] In this embodiment, a convolutional neural network (CNN1) can be used for feature extraction. The feature extraction process can be described by formula (2), which is shown below.

[0197] F t =CNN1(I′) t ;θ CNN1 (2);

[0198] Among them, F t θ represents the image features extracted from the image (called the initial image features). CNN1 This represents the model parameters of the CNN1 convolutional neural network.

[0199] It's important to note that a convolutional neural network (CNN) is a deep learning model that effectively extracts features from images by mimicking the local perception and hierarchical structure of the human visual system. Typically, a CNN includes an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. The input layer receives the raw image data, usually a three-dimensional tensor with height, width, and depth (depth of 1 for grayscale images and 3 for color images, corresponding to the red, green, and blue channels, respectively). The convolutional layer contains a set of learnable convolutional kernels (or filters). Each kernel slides across the input image (convolution operation), generating a set of feature maps. The convolution operation can extract local features of the image, such as edges and corners. Pooling layers are used to downsample feature maps, reducing the dimensionality of the data while retaining effective information. Common pooling methods include max pooling and average pooling. Fully connected layers are used to flatten multidimensional feature maps into one-dimensional vectors, similar to traditional neural networks. They are usually located in the last layer of convolutional neural networks and are used for the final classification or regression tasks. Output layers are used to select the appropriate output layer according to the task type.

[0200] The embodiments of this application use an attention mechanism for weight allocation; the weight allocation process can be described by formula (3), which is shown below.

[0201] A t =Softmax1(w1·F t +b) (3);

[0202] Among them, A t Represented as image feature F t The attention weights are assigned, w1 represents the weight assignment parameters, b represents the weight assignment bias term, and Softmax1 is the activation function used to ensure A t The sum of all weights is 1.

[0203] In this embodiment, feature determination is performed after attention weights are assigned to the extracted image features. That is, by fusing image features and attention weights, the features to be detected for clipping can be obtained. The process of fusing image features and attention weights can be described by formula (4), which is shown below.

[0204] F′ t =A t ⊙F t (4);

[0205] Among them, F′t The fused features are the features to be detected (referred to as the image features to be detected); ⊙ represents the multiplication between elements, used to fuse image features and attention weights.

[0206] In this embodiment of the application, the feature F′ to be detected for penetration is... t Perform clipping detection to obtain the corresponding clipping detection results. The feature F′ to be detected is... t The process of performing the penetration detection can be described by formula (5), which is shown below.

[0207] C t =Softmax2(w2·F t ′+c) (5);

[0208] Among them, C t w2 represents the first clipping detection result (referred to as the first clipping detection result) obtained based on the features to be detected, c represents the clipping detection bias term, and Softmax2 is the activation function.

[0209] It should be noted that by analyzing each video frame image through the attention mechanism, features in the video frame image that are helpful for clipping detection can be identified; thus, the accuracy of clipping detection can be improved, as well as the processing speed of clipping detection can be increased.

[0210] In this embodiment, the process of detecting clipping based on the attention mechanism of the preprocessed results can be implemented by a trained neural network model. The training process of the neural network model is iterative and is implemented by batch training datasets. The training datasets include video frame images with clipping and video frame images without clipping. As training progresses, the neural network model will gradually learn to ignore information unrelated to clipping detection and enhance information related to clipping detection.

[0211] Next, the process of performing clipping detection on the preprocessed results based on semantic information is explained.

[0212] It should be noted that semantic information is obtained by performing semantic analysis on either the unprocessed or preprocessed video frame images (here, we will use the preprocessed video frame images as an example). Semantic analysis is used to interpret the content in the image to obtain each element in the image and the semantic relationships between the elements. Here, image processing and natural language processing can be used for semantic analysis.

[0213] In this embodiment of the application, when performing semantic analysis, image processing is first used to extract semantic features. The extraction of semantic features can be achieved through a deep learning model, such as a convolutional neural network. Here, the extraction of semantic features can be described by formula (6), which is shown below.

[0214] S t =CNN2(I′) t ;θ CNN2 (6);

[0215] Among them, S t θ represents the semantic features extracted from the image (called image semantic features). CNN2 This represents the model parameters of the convolutional neural network CNN2.

[0216] Here, the process of obtaining semantic information based on semantic features can be described by formula (7), which is shown below.

[0217] S′ t =RNN(S) t ;θ RNN (7);

[0218] Among them, S′ t θ represents semantic information obtained based on semantic features (referred to as image semantic information). RNN This represents the model parameters of a recurrent neural network (RNN).

[0219] It should be noted that recurrent neural networks (RNNs) are suitable for processing sequential data, such as time series analysis and natural language processing. Therefore, this embodiment uses an RNN to obtain semantic information from video frame images. An RNN typically includes an input layer, a hidden layer, and an output layer. The input layer receives the input of the sequential data at each time step. The hidden layer includes multiple neurons, each taking the input of the current time step and the hidden state of the previous time step as input. Thus, the output of the hidden layer depends not only on the input of the current time step but also on the inputs of all previous time steps (passed through the hidden states). The output layer generates the prediction result for each time step based on the output of the hidden layer.

[0220] After obtaining the semantic information, a graph network is used to process the semantic information in order to model the semantic relationships between the elements in the image, as shown in formula (8).

[0221] G t =Graph(S′) t ;θ Graph (8);

[0222] Among them, G t θ is the output of a graph network used to represent the semantic relationships between elements in an image (called graph network features). Graph These are the parameters of the graph network.

[0223] After obtaining the graph network output, clipping detection is performed on the graph network output; here, the process of clipping detection on the graph network output can be described by formula (9), as shown below.

[0224] C′ t =Softmax3(w3·G t +d) (9);

[0225] Among them, C′ t The output of the graph network represents the clipping detection result (called the second clipping detection feature), w3 represents the clipping detection weight, d represents the clipping detection bias term, and Softmax3 is the activation function.

[0226] It is understandable that semantic analysis improves the accurate understanding of image content and realizes the semantic relationships between various elements in the image. Therefore, when semantic analysis is used for clipping detection in game scenes, it can accurately understand the rules and logic of the game world, as well as the motion attributes of characters and objects, thereby improving the accuracy of clipping detection.

[0227] Finally, the process of obtaining the penetration detection results is explained.

[0228] It should be noted that, in conjunction with the results of the penetration testing C t And the result of the penetration test C′ t The process of obtaining the penetration detection result can be described by formula (10), which is shown below.

[0229] D t =C t ∩C′ t (10);

[0230] Among them, D t This represents the final penetration detection result (referred to as the target penetration detection result); here, when 0 indicates no penetration and 1 indicates penetration, the penetration detection result C... t And the result of the penetration test C′ t When both values ​​are 1, the final result of the penetration detection is D. t Only then does it indicate the existence of clipping.

[0231] For example, see Figure 12 , Figure 12This is an exemplary schematic diagram of the penetration detection result provided in an embodiment of this application; as shown... Figure 12 As shown, for video frame image 12-1, since there are three instances of clipping (positions 12-21, 12-22, and 12-23) in video frame image 12-1, the clipping detection result C is... t And the result of the penetration test C′ t All are 1.

[0232] Understandably, in game scenarios, performing clipping detection frame-by-frame on recorded videos of 3D game models allows for the precise capture of every possible clipping moment, improving the comprehensiveness and precision of clipping detection. Processing image features through an attention mechanism enables accurate and rapid identification of features relevant to clipping detection, thereby enhancing the accuracy and efficiency of clipping detection. Furthermore, semantic analysis of images allows for accurate processing of the semantic relationships between various elements within the image, further improving the accuracy of clipping detection. In addition, the information detection method provided in this application embodiment can automatically and intelligently perform batch clipping detection, improving the intelligence, accuracy, and efficiency of clipping detection.

[0233] The following description continues to illustrate the exemplary structure of the information detection device 455 provided in the embodiments of this application as a software module. In some embodiments, such as... Figure 2 As shown, the software module stored in the information detection device 455 of the memory 450 may include:

[0234] The feature processing module 4551 is used to perform attention processing on the initial image features of the image to be detected to obtain the image features to be detected, wherein the image to be detected is an image of a virtual object model;

[0235] The first detection module 4552 is used to perform clipping detection on the features of the image to be detected and obtain a first clipping detection result.

[0236] Network construction module 4553 is used to construct an element graph network based on the image semantic information of the image to be detected, wherein the element graph network represents the semantic relationship between each image element in the image to be detected;

[0237] The second detection module 4554 is used to perform clipping detection on the element graph network and obtain a second clipping detection result.

[0238] The result acquisition module 4555 is used to combine the first clipping detection result and the second clipping detection result to obtain the target clipping detection result of the image to be detected.

[0239] In this embodiment of the application, the feature processing module 4551 is further configured to perform attention processing on the initial image features of the image to be detected to obtain the attention weights of the initial image features; and to fuse the initial image features and the attention weights to obtain the image features to be detected.

[0240] In this embodiment of the application, the semantic relationship includes at least one of the following: hierarchical relationship, positional relationship, and action relationship; the hierarchical relationship represents the display hierarchy between the various image elements, the positional relationship represents the relative position between the various image elements, and the action relationship represents the rigid body motion action between the various image elements.

[0241] In this embodiment of the application, the network construction module 4553 is further configured to extract the image semantic features of the image to be detected; and determine the image semantic information of the image to be detected based on the image semantic features.

[0242] In this embodiment of the application, the network construction module 4553 is further configured to determine each of the image elements of the image to be detected and the semantic relationship between each of the image elements from the image semantic information; and to construct the element graph network by using each of the image elements as graph network nodes and constructing graph network node edges based on the semantic relationship.

[0243] In this embodiment of the application, the feature processing module 4551 is further configured to record video of the process of executing motion instructions on the virtual object model to obtain a video to be detected; divide the video to be detected into video frames to obtain a video frame image sequence; traverse the video frame image sequence and determine the traversed video frame image as the image to be detected.

[0244] In this embodiment of the application, the second detection module 4554 is further configured to perform clipping detection on the element graph network to obtain a third clipping detection result; obtain N fourth clipping detection results corresponding to the first N video frames of the image to be detected, where N is a positive integer; and combine the third clipping detection result and the N fourth clipping detection results to obtain the second clipping detection result.

[0245] In this embodiment of the application, the feature processing module 4551 is further configured to perform image preprocessing on the image to be detected to obtain an image to be processed, wherein the image preprocessing includes at least one of the following: noise reduction, brightness adjustment and contrast adjustment; and to extract features from the image to be processed to obtain the initial image features.

[0246] In this embodiment of the application, the result acquisition module 4555 is further configured to: when both the first and second clipping detection results indicate that the virtual object model is clipping, determine the information in the image to be detected that the virtual object model is clipping as the target clipping detection result; when both the first and second clipping detection results indicate that the virtual object model is not clipping, determine the information in the image to be detected that the virtual object model is not clipping as the target clipping detection result; and when one of the first and second clipping detection results indicates that the virtual object model is clipping, determine the information in the image to be detected that the virtual object model is to be clipped and verified as the target clipping detection result.

[0247] In this embodiment, the first clipping detection result is achieved through a first clipping detection model; the information detection device 455 further includes a model training module 4556, used to acquire a first image sample and a first clipping annotation result of the first image sample; to perform clipping detection on the first image sample using a first model to be trained, to obtain a first clipping prediction result, wherein the first model to be trained is a neural network model to be trained for clipping detection based on attention processing; and to train the first model to be trained based on the difference between the first clipping prediction result and the first clipping annotation result, to obtain the first clipping detection model.

[0248] In this embodiment, the second clipping detection result is achieved through a second clipping detection model; the model training module 4556 is used to acquire a second image sample and a second clipping annotation result of the second image sample; the second image sample is subjected to clipping detection using a second model to be trained to obtain a second clipping prediction result, wherein the second model to be trained is a neural network model to be trained for clipping detection based on semantic information; the second model to be trained is trained based on the difference between the second clipping prediction result and the second clipping annotation result to obtain the second clipping detection model.

[0249] In this embodiment of the application, the information detection device 455 further includes a model reconstruction module 4557, which is used to reconstruct the virtual object model based on the image to be detected when the target clipping detection result indicates that the virtual object model exists in the image to be detected, to obtain a virtual object reconstruction model; to record the process of executing motion commands on the virtual object reconstruction model to obtain a target image corresponding to the content of the image to be detected; and to perform clipping detection on the target image to obtain a new clipping detection result.

[0250] This application provides a computer program product, which includes computer-executable instructions or a computer program stored in a computer-readable storage medium. The processor of an electronic device reads the computer-executable instructions or computer program from the computer-readable storage medium and executes the computer-executable instructions or computer program, causing the electronic device to perform the information detection method described in this application.

[0251] This application provides a computer-readable storage medium storing computer-executable instructions or a computer program. When the computer-executable instructions or the computer program are executed by a processor, the processor will execute the information detection method provided in this application. For example, ... Figure 3 The information detection method shown.

[0252] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0253] In some embodiments, computer-executable instructions may take the form of programs, software, software modules, scripts, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as stand-alone programs or as modules, components, subroutines, or other units suitable for use in a computing environment.

[0254] As an example, computer-executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple co-located files (e.g., files that store one or more modules, subroutines, or code sections).

[0255] As an example, computer-executable instructions can be deployed to execute on a single electronic device (in which case, this single electronic device is an information detection device), or to execute on multiple electronic devices located at one location (in which case, the multiple electronic devices located at one location are information detection devices), or to execute on multiple electronic devices distributed across multiple locations and interconnected via a communication network (in which case, the multiple electronic devices distributed across multiple locations and interconnected via a communication network are information detection devices).

[0256] It is understood that in the embodiments of this application, data related to virtual object models and images to be detected are involved. When the embodiments of this application are applied to specific products or technologies, the permission or consent of the information subject is required, and the collection, use and processing of related data must comply with relevant laws, regulations and standards, and conform to the principles of legality, legitimacy and necessity. It does not involve obtaining data types prohibited or restricted by laws and regulations, and will not hinder the normal operation of the target website.

[0257] In summary, this application's embodiments, targeting the image to be detected from a virtual object model, combine attention processing and image semantic information for clipping detection, achieving a multi-dimensional automated clipping detection solution for the image to be detected from a virtual object model. This improves the accuracy and efficiency of clipping detection. Furthermore, in a virtual scene, by performing clipping detection frame-by-frame on the recorded video of the virtual object model, every possible clipping moment can be precisely captured, enhancing the comprehensiveness and precision of clipping detection.

[0258] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. An information detection method characterized by comprising: The method comprises: performing attention processing on initial image features of a to-be-detected image to obtain to-be-detected image features, the to-be-detected image being an image of a virtual object model; performing pass-through detection on the to-be-detected image features to obtain a first pass-through detection result; constructing an element graph network based on image semantic information of the to-be-detected image, the element graph network representing semantic relationships between image elements in the to-be-detected image; performing pass-through detection on the element graph network to obtain a second pass-through detection result; combining the first pass-through detection result and the second pass-through detection result to obtain a target pass-through detection result of the to-be-detected image.

2. The method of claim 1, wherein, The method further comprises: performing attention processing on initial image features of a to-be-detected image to obtain to-be-detected image features, the to-be-detected image being an image of a virtual object model; performing attention processing on the initial image features of the to-be-detected image to obtain attention weights of the initial image features; 3. The method of claim 1, wherein, combining the initial image features and the attention weights to obtain the to-be-detected image features.

4. The method according to any one of claims 1 to 3, characterized in that, The semantic relationships comprise at least one of a hierarchical relationship, a positional relationship, and an action relationship; the hierarchical relationship represents display hierarchies between the image elements, the positional relationship represents relative positions between the image elements, and the action relationship represents rigid motion actions between the image elements. The method further comprises: extracting image semantic features of the to-be-detected image; 5. The method according to claim 1 or 3, characterized in that, determining image semantic information of the to-be-detected image based on the image semantic features. The method further comprises: determining, from the image semantic information, the image elements of the to-be-detected image and the semantic relationships between the image elements; 6. The method of claim 1, wherein, constructing the element graph network by taking the image elements as graph network nodes and constructing graph network node edges based on the semantic relationships. The method further comprises: recording a video of a process of executing a motion instruction on the virtual object model to obtain a to-be-detected video; dividing video frames of the to-be-detected video to obtain a sequence of video frame images; 7. The method according to any of claims 1 to 3, 6, characterized in that, traversing the sequence of video frame images and determining a video frame image that is traversed as a to-be-detected image. The method further comprises: performing pass-through detection on the element graph network to obtain a third pass-through detection result; obtaining N fourth pass-through detection results corresponding to N previous video frame images of the to-be-detected image, N being a positive integer; 8. The method according to any of claims 1 to 3, 6, characterized in that, combining the third pass-through detection result and the N fourth pass-through detection results to obtain the second pass-through detection result. The method further comprises: performing image preprocessing on the to-be-detected image to obtain a to-be-processed image, the image preprocessing comprising at least one of denoising, brightness adjustment, and contrast adjustment; Feature extraction is performed on the image to be processed to obtain the initial image features.

9. The method according to any of claims 1 to 3, 6, characterized in that, The first penetration detection result and the second penetration detection result are combined to obtain a target penetration detection result of the image to be detected, including: When the first penetration detection result and the second penetration detection result both indicate that the virtual object model penetrates, information that the virtual object model penetrates in the image to be detected is determined as the target penetration detection result; When the first penetration detection result and the second penetration detection result both indicate that the virtual object model does not penetrate, information that the virtual object model does not penetrate in the image to be detected is determined as the target penetration detection result; When one of the first penetration detection result and the second penetration detection result indicates that the virtual object model penetrates, information that the virtual object model needs to be detected for penetration in the image to be detected is determined as the target penetration detection result.

10. The method according to any of claims 1 to 3, 6, characterized in that, The first penetration detection result is achieved by a first penetration detection model; The first penetration detection model is obtained by training the following steps: Obtain a first image sample and a first penetration label result of the first image sample; Perform penetration detection on the first image sample by using a first to-be-trained model to obtain a first penetration prediction result, the first to-be-trained model being a neural network model to be trained for penetration detection based on attention processing; Train the first to-be-trained model based on the difference between the first penetration prediction result and the first penetration label result to obtain the first penetration detection model.

11. The method according to any of claims 1 to 3, 6, characterized in that, The second penetration detection result is achieved by a second penetration detection model; The second penetration detection model is obtained by training the following steps: Obtain a second image sample and a second penetration label result of the second image sample; Perform penetration detection on the second image sample by using a second to-be-trained model to obtain a second penetration prediction result, the second to-be-trained model being a neural network model to be trained for penetration detection based on semantic information; Train the second to-be-trained model based on the difference between the second penetration prediction result and the second penetration label result to obtain the second penetration detection model.

12. The method according to any of claims 1 to 3, 6, characterized in that, After the first penetration detection result and the second penetration detection result are combined to obtain the target penetration detection result of the image to be detected, the method further includes: When the target penetration detection result indicates that the virtual object model penetrates in the image to be detected, perform model reconstruction on the virtual object model based on the image to be detected to obtain a virtual object reconstructed model; Video record the process of executing a motion instruction on the virtual object reconstructed model to obtain a target image corresponding to the content of the image to be detected; Perform penetration detection on the target image to obtain a new penetration detection result.

13. An information detection device, characterized by comprising: The information detection device includes: A feature processing module configured to perform attention processing on initial image features of an image to be detected to obtain image features of the image to be detected, the image to be detected being an image of a virtual object model; The first detection module is configured to perform a pass-through detection on the image feature to be detected to obtain a first pass-through detection result. The network construction module is configured to construct an element graph network based on image semantic information of the image to be detected, the element graph network representing semantic relationships between image elements in the image to be detected. The second detection module is configured to perform a pass-through detection on the element graph network to obtain a second pass-through detection result. The result acquisition module is configured to combine the first pass-through detection result and the second pass-through detection result to obtain a target pass-through detection result of the image to be detected.

14. An electronic device for information detection, characterized by, The electronic device comprises: a memory configured to store computer executable instructions or computer programs; a processor configured to execute the computer executable instructions or computer programs stored in the memory to implement the information detection method in any one of claims 1 to 12.

15. A computer-readable storage medium storing computer-executable instructions or a computer program, wherein the computer-executable instructions or the computer program comprise the steps of: The computer executable instructions or computer programs are executed by the processor to implement the information detection method in any one of claims 1 to 12. ​ 16. A computer program product comprising computer-executable instructions or a computer program, characterized in that, The computer executable instructions or computer programs are executed by the processor to implement the information detection method in any one of claims 1 to 12. The computer executable instructions or computer programs are executed by the processor to implement the information detection method in any one of claims 1 to 12.

Images