Label positioning method, device, computer equipment and storage medium for question semantic graph

Abstract

The present invention discloses a method, device, computer equipment and storage medium for label positioning of a problem semantic graph, including: obtaining a file to be positioned fed back by a user, wherein the file to be positioned includes a rendering task ID and a camera ID; obtaining a corresponding rendering task result according to the rendering task ID, determining a problem semantic graph and a first mapping relationship between the problem semantic graph and an entity ID from the rendering task result according to the camera ID; visualizing the problem semantic graph after coloring, determining problem pixels from the visualization result, determining the entity ID corresponding to the problem pixels according to the first mapping relationship between the problem pixels, the problem semantic graph and the entity ID; and determining the label corresponding to the entity ID from a three-dimensional scene data set according to the entity ID. The method and device can realize fast and accurate label positioning of the problem semantic graph.

Description

Technical Field

[0001] The present invention belongs to the field of image processing, and in particular relates to a label positioning method, device, computer equipment and storage medium for a question semantic graph. Background Art

[0002] In recent years, artificial intelligence technology has developed rapidly, and its commercialization speed has exceeded expectations. Artificial intelligence will bring disruptive changes to the entire society and has become an important development strategy for countries in the future. The evolution of algorithms with deep learning as the core, with its super evolutionary ability, can solve various problems through training and construction of large-scale convolutional neural networks similar to the structure of the human brain with the support of big data. The development of big data, especially the emergence of massive data, has broken the traditional way of deep learning from big data and brought about disruptive changes in various fields.

[0003] In the process of model training, sample data and its semantic labels are needed. The massive unstructured data obtained through various means contains rich and valuable semantic information. To extract this semantic information, a lot of manual annotation and analysis is required. Due to the low efficiency and high cost of manual annotation, manual semantic annotation of massive data is not realistic in practical applications.

[0004] With the rapid development of artificial intelligence and breakthroughs in deep learning technology, computer vision technology based on deep learning is becoming more mature. Researchers are trying to make machines understand more complex semantic information in visual information. Therefore, image semantic description research has emerged in the intersection of computer vision and natural language processing. The goal of image semantic description technology is to achieve the conversion of images in visual space to text descriptions in semantic space. The method realizes the mapping from images to text description sentences, gives semantic interpretation to visual data, and realizes the mapping from visual space to semantic space.

[0005] In response to the demand for sample data in the era of artificial intelligence, some data generation platforms obtain a large number of semantic maps by changing the camera position to roam and render in a three-dimensional scene. These semantic maps have semantic labels and can be used as sample data. These data generation platforms are open to all users. Users generally use the data generation platform to obtain the required semantic maps, but some semantic maps may have some problems and need to be fed back to the platform for repositioning of semantic labels.

[0006] The problem information reported by users basically includes the task ID, the camera ID corresponding to the problem semantic graph, and the problem manifestation, that is, what specific problems have occurred. The verification personnel of the data generation platform generally conduct offline verification based on the feedback problem information. Offline verification requires manual downloading and decompression of multiple files, and requires running multiple scripts locally, which makes the label verification and positioning process extremely inefficient. Summary of the invention

[0007] In view of the above, an object of the present invention is to provide a method, apparatus, computer device and storage medium for locating labels of a question semantic graph, so as to achieve fast and accurate positioning of labels of the question semantic graph.

[0008] To achieve the above-mentioned object of the invention, a first embodiment provides a problem location method of a semantic graph, comprising the following steps:

[0009] Obtain the file to be located reported by the user, wherein the file to be located includes a rendering task ID, a camera ID, and problem data;

[0010] Obtain a corresponding rendering task result according to the rendering task ID, determine a problem semantic graph from the rendering task result according to the camera ID, and a first mapping relationship between the problem semantic graph and the entity ID;

[0011] The problem semantic graph is visualized after coloring, the problem pixel is determined from the visualization result combined with the problem data, and the entity ID corresponding to the problem pixel is determined according to the first mapping relationship between the problem semantic graph and the entity ID;

[0012] The label corresponding to the entity ID is determined from the three-dimensional scene data set according to the entity ID.

[0013] To achieve the above-mentioned object of the invention, a second embodiment provides a label positioning method for a question semantic graph, comprising the following steps:

[0014] Obtain the file to be located reported by the user, wherein the file to be located includes a rendering task ID, a camera ID, and problem data;

[0015] Obtain the corresponding rendering intermediate result according to the rendering task ID, determine the problem model graph corresponding to the problem semantic graph from the rendering intermediate result according to the camera ID, and the second mapping relationship between the problem model graph and the entity ID;

[0016] The problem model graph is visualized after coloring, the problem pixel is determined from the visualization result combined with the problem data, and the entity ID corresponding to the problem pixel is determined according to the second mapping relationship between the problem model graph and the entity ID;

[0017] The label corresponding to the entity ID is determined from the three-dimensional scene data set according to the entity ID.

[0018] In one embodiment, the problem data includes problem pixels, and then the entity ID corresponding to the problem pixels is determined based on the second mapping relationship between the problem model graph and the entity ID, or the entity ID corresponding to the problem pixel is determined based on the second mapping relationship between the problem model graph and the entity ID.

[0019] In one embodiment, obtaining the corresponding rendering task result according to the rendering task ID includes:

[0020] Get the resource link of the rendering task result according to the rendering task ID, download and parse the json file through the resource link, and get the rendering task result.

[0021] In one embodiment, obtaining the corresponding rendering intermediate result according to the rendering task ID includes:

[0022] Get the resource link of the rendering intermediate result according to the rendering task ID, download and parse the json file through the resource link, and get the rendering intermediate result.

[0023] In one embodiment, the label positioning method further includes: generating a problem file according to the labels corresponding to the problem pixels in the problem semantic graph and supporting downloading, wherein the problem file includes each pixel value in the problem semantic graph and the corresponding label value distribution.

[0024] To achieve the above-mentioned object of the invention, a third aspect of the embodiment provides a label positioning device for a question semantic graph, comprising:

[0025] A file acquisition module, used to acquire the to-be-located file fed back by the user, wherein the to-be-located file includes a rendering task ID, a camera ID, and problem data;

[0026] A mapping relationship determination module, used to obtain a corresponding rendering task result according to a rendering task ID, determine a problem semantic graph from the rendering task result according to a camera ID, and a first mapping relationship between the problem semantic graph and an entity ID;

[0027] An entity ID determination module is used to visualize the problem semantic graph after coloring, determine the problem pixel point from the visualization result combined with the problem data, and determine the entity ID corresponding to the problem pixel point according to the first mapping relationship between the problem semantic graph and the entity ID;

[0028] The label determination module is used to determine the label corresponding to the entity ID from the three-dimensional scene data set according to the entity ID.

[0029] To achieve the above-mentioned purpose of the invention, a fourth aspect of the embodiment provides a label positioning device for a question semantic graph, comprising:

[0030] A file acquisition module, which acquires the file to be located reported by the user, wherein the file to be located includes a rendering task ID, a camera ID, and problem data;

[0031] A mapping relationship determination module, used to obtain a corresponding rendering task result according to a rendering task ID, determine a problem semantic graph from the rendering task result according to a camera ID, and a first mapping relationship between the problem semantic graph and an entity ID;

[0032] An entity ID determination module is used to visualize the problem model graph after coloring, determine the problem pixel point from the visualization result combined with the problem data, and determine the entity ID corresponding to the problem pixel point according to the second mapping relationship between the problem model graph and the entity ID;

[0033] The label determination module is used to determine the label corresponding to the entity ID from the three-dimensional scene data set according to the entity ID.

[0034] To achieve the above-mentioned purpose of the invention, the fifth aspect embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above-mentioned method for label positioning of the question semantic graph when executing the computer program.

[0035] To achieve the above-mentioned purpose of the invention, the sixth aspect embodiment provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is processed and executed, the steps of the above-mentioned method for label positioning of the question semantic graph are implemented.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The label positioning method, device, computer equipment and storage medium of the problem semantic graph provided in the embodiment obtain the mapping relationship between the problem semantic graph and the entity ID from the rendering task result and the rendering intermediate result based on the rendering task ID and camera ID fed back by the user, and then determine the entity ID corresponding to the problem pixel based on the mapping relationship, and finally determine the label corresponding to the entity ID from the three-dimensional scene data set based on the entity ID. The process is completed automatically and does not rely on the person locating the problem to master the code debugging ability and the problem troubleshooting process to easily locate the problem, which facilitates the subsequent cross-group docking and problem handling, thereby making the entire process smoother and more efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0039] Figure 1 It is a flow chart of a label positioning method of a question semantic graph provided by an embodiment;

[0040] Figure 2 is a flowchart of a label positioning method for a question semantic graph provided by another embodiment;

[0041] Figure 3It is a structural schematic diagram of a label positioning device for a question semantic graph provided by an embodiment. DETAILED DESCRIPTION

[0042] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation methods described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0043] Data generation platforms such as EUS (Environment UnderStanding Solution) can output rendering images, reverse images, depth images, semantic images, example images, etc. based on 3D scene data sets, and have their own label data with content information for users to use.

[0044] When the data generation platform outputs the graph, there will be a problem of unclear semantic graph labels, among which unclear labels can be divided into three categories. One is that the label of the semantic graph has an unrecognized value, that is, it is shown that part of the black graph is white. The second category is that the label of the semantic graph is inaccurate, that is, it is actually the label of object A that is marked on object B. For example, in the field of home decoration design, the label of the wall is actually marked on the ground. The third category is that there are many overlapping object models after the semantic graph is colored, which is manifested as a lot of noise at the junction of multiple object models after the semantic graph is colored. Regardless of which type of problem, when locating the problem, it is necessary to parse the frame of the problem picture, look at the label value of the problem pixel point, and then go to the three-dimensional model and the three-dimensional scene data set for comparison to further determine the node where the problem occurs. However, this process is manually operated, various parsing and script re-running, and the link is long and time-consuming, resulting in low efficiency and inaccurate problems. To solve this problem, the embodiment provides a label positioning method and device for a problem semantic graph, which realizes the rapid and accurate positioning of the label of the problem semantic graph in an automated manner.

[0045] Figure 1 FIG. 1 is a flow chart of a method for label positioning of a question semantic graph provided by an embodiment. Figure 1 As shown, the label positioning method of the question semantic graph provided in the embodiment includes the following steps:

[0046] S101, obtaining a file to be located reported by a user, wherein the file to be located includes a rendering task ID, a camera ID, and problem data.

[0047] When users use the data generation platform to generate semantic graphs, they will feedback the files to be located containing problem information based on the generated results, where the files to be located contain the rendering task ID and camera ID. The rendering task ID is used to locate the rendering task to facilitate the search for the final result data of the rendering task and the intermediate result data of the rendering. Since the data generation platform obtains a large number of semantic graphs through mobile rendering, the output semantic graph is a continuous frame image, and the camera ID fed back by the user can accurately locate the problem semantic graph. Furthermore, compared with directly feeding back the problem semantic graph, it is easier to determine the data of each link of the rendering task by feeding back the rendering task ID and camera ID, which is convenient for tracing.

[0048] In the embodiment, the question data illustrates the question surface of the semantic graph, which specifically includes the above three types of questions.

[0049] S102, obtaining a corresponding rendering task result according to the rendering task ID, determining a problem semantic graph from the rendering task result according to the camera ID, and a first mapping relationship between the problem semantic graph and the entity ID.

[0050] In an embodiment, obtaining the corresponding rendering task result according to the rendering task ID includes: obtaining a resource link of the rendering task result according to the rendering task ID, downloading and parsing a json file through the resource link, and obtaining the rendering task result, that is, downloading a compressed package zip through the resource link and decompressing and formatting the display json file to obtain the rendering task result. The rendering task result includes a semantic graph of consecutive frames, and a mapping relationship between the semantic graph and the entity ID in the three-dimensional scene data.

[0051] The rendered output semantic map is obtained by semantically segmenting the image, that is, automatically segmenting and identifying the content in the image and giving the content a semantic label. For example, for a room image, the corresponding semantic map will distinguish between walls, ceilings, floors, beds, etc., generally by color, and each color corresponds to a semantic label.

[0052] After obtaining the continuous frame semantic graph and the mapping relationship between the semantic graph and the entity ID in the three-dimensional scene data, the semantic graph corresponding to the camera ID is found as the problem semantic graph according to the camera ID fed back by the user, and the first mapping relationship between the problem semantic graph and the entity ID is also found. The first mapping relationship is used to determine the entity ID corresponding to the problem pixel.

[0053] S103, coloring and visualizing the question semantic graph, determining the problem pixel point from the visualization result in combination with the problem data, and determining the entity ID corresponding to the problem pixel point according to the first mapping relationship between the question semantic graph and the entity ID.

[0054] The problem semantic map generated by the data generation platform is invisible to the naked eye and needs to be visualized by coloring it. Then, a drawing tool is used to determine the problem pixel points from the visualization results combined with the problem data. Finally, the entity ID corresponding to the problem pixel point is determined based on the first mapping relationship.

[0055] In another embodiment, the problem data fed back by the user directly includes the problem pixel points. In this case, there is no need to color and visualize the problem semantic graph to determine the pixel points. That is, the entity ID corresponding to the problem pixel point can be directly determined based on the first mapping relationship between the problem semantic graph and the entity ID.

[0056] S104, determining a label corresponding to the entity ID from the three-dimensional scene data set according to the entity ID.

[0057] During the rendering process, a three-dimensional scene data set is also obtained. The three-dimensional scene data set includes the entity ID in the three-dimensional scene data set, the label corresponding to the entity, the entity size, the entity position, the entity material, and the rendering light. After the entity ID corresponding to the problem pixel is determined, the corresponding label value is found from the three-dimensional scene data set based on the entity ID. The label value is the correct label value of the problem pixel and is displayed.

[0058] S105, generating a question file based on the labels corresponding to the question pixels in the question semantic graph and supporting downloading.

[0059] In the embodiment, after determining the label corresponding to the problem pixel in the problem semantic map, a problem file is generated, which can be a .exr file and supports downloading. The problem file includes each pixel value and the corresponding label value distribution in the problem semantic map. After the user downloads the problem file, he opens the problem file through the offline tev tool to see each pixel value and the corresponding label value distribution in the problem semantic map.

[0060] Figure 2 FIG. 1 is a flow chart of a method for label positioning of a question semantic graph provided by another embodiment. Figure 2 As shown, the label positioning method of the question semantic graph provided by the embodiment includes:

[0061] S201, obtaining a file to be located reported by a user, wherein the file to be located includes a rendering task ID, a camera ID, and problem data.

[0062] The process of obtaining the file to be located in S201 is the same as that of obtaining the file to be located in S101, and will not be repeated here.

[0063] S202, obtaining the corresponding rendering intermediate result according to the rendering task ID, determining the problem model graph corresponding to the problem semantic graph from the rendering intermediate result according to the camera ID, and the second mapping relationship between the problem model graph and the entity ID.

[0064] In an embodiment, obtaining the corresponding rendering intermediate result according to the rendering task ID includes: obtaining the resource link of the rendering intermediate result according to the rendering task ID, downloading and parsing the json file through the resource link, and obtaining the rendering intermediate result, that is, downloading the compressed package zip through the resource link and decompressing and formatting the display json file to obtain the rendering intermediate result. The rendering intermediate result includes the model diagram of the continuous frame, and the mapping relationship between the model diagram and the entity ID in the three-dimensional scene data.

[0065] Model graph refers to the model segmentation graph output by the rendering engine (such as FF engine), which is used to distinguish models such as furniture and walls and mark the location of the 3D scene where the model is located. Each model graph is semantically related to the output Figure 1 Based on this, after obtaining the model graph of the continuous frames and the mapping relationship between the model graph and the entity ID in the 3D scene data, the problem model graph corresponding to the problem semantic graph corresponding to the camera ID is found according to the camera ID fed back by the user, and the second mapping relationship between the problem model graph and the entity ID is also found. The second mapping relationship is used to determine the entity ID corresponding to the problem pixel.

[0066] S203, coloring and visualizing the problem model graph, determining the problem pixel from the visualization result in combination with the problem data, and determining the entity ID corresponding to the problem pixel based on the second mapping relationship between the problem model graph and the entity ID.

[0067] The problem model diagram generated by the data generation platform is also invisible to the naked eye and needs to be visualized by coloring it. Then, a drawing tool is used to determine the problem pixel points from the visualization results combined with the problem data. Finally, the entity ID corresponding to the problem pixel point is determined based on the second mapping relationship.

[0068] In another embodiment, the problem data fed back by the user directly includes the problem pixels. In this case, there is no need to color and visualize the problem model graph to determine the pixels. That is, the entity ID corresponding to the problem pixel can be directly determined based on the second mapping relationship between the problem coloring graph and the entity ID.

[0069] S204: Determine a label corresponding to the entity ID from the three-dimensional scene data set according to the entity ID.

[0070] The process of determining the label corresponding to the entity ID from the three-dimensional scene data set according to the entity ID in S204 is the same as S104, and will not be repeated here.

[0071] S205, generating a question file based on the labels corresponding to the question pixels in the question semantic graph and supporting downloading.

[0072] The process of generating a question file based on the labels corresponding to the question pixels in the question semantic graph and supporting downloading in S205 is the same as S105 and will not be repeated here.

[0073] Figure 3 FIG. 1 is a schematic diagram of a structure of a label positioning device for a question semantic graph provided by an embodiment. Figure 3 As shown, the label positioning device of the question semantic graph provided by the embodiment includes:

[0074] A file acquisition module, used to acquire the to-be-located file fed back by the user, wherein the to-be-located file includes a rendering task ID, a camera ID, and problem data;

[0075] A mapping relationship determination module, used to obtain a corresponding rendering task result according to a rendering task ID, determine a problem semantic graph from the rendering task result according to a camera ID, and a first mapping relationship between the problem semantic graph and an entity ID;

[0076] An entity ID determination module is used to visualize the problem semantic graph after coloring, determine the problem pixel point from the visualization result combined with the problem data, and determine the entity ID corresponding to the problem pixel point according to the first mapping relationship between the problem semantic graph and the entity ID;

[0077] The label determination module is used to determine the label corresponding to the entity ID from the three-dimensional scene data set according to the entity ID.

[0078] The label positioning device for the question semantic graph provided in the embodiment further includes:

[0079] A file acquisition module, which acquires the file to be located reported by the user, wherein the file to be located includes a rendering task ID, a camera ID, and problem data;

[0080] A mapping relationship determination module, used to obtain a corresponding rendering task result according to a rendering task ID, determine a problem semantic graph from the rendering task result according to a camera ID, and a first mapping relationship between the problem semantic graph and an entity ID;

[0081] An entity ID determination module is used to visualize the problem model graph after coloring, determine the problem pixel point from the visualization result combined with the problem data, and determine the entity ID corresponding to the problem pixel point according to the second mapping relationship between the problem model graph and the entity ID;

[0082] The label determination module is used to determine the label corresponding to the entity ID from the three-dimensional scene data set according to the entity ID.

[0083] It should be noted that the label positioning device for the question semantic graph provided in the above embodiment should be illustrated by the division of the above functional modules when performing label positioning of the question semantic graph. The above functional allocation can be completed by different functional modules as needed, that is, the internal structure of the terminal or server is divided into different functional modules to complete all or part of the functions described above. Figure 1 and Figure 2 The embodiment of the label positioning method of the problem semantic graph shown in FIG. 1 belongs to the same concept, and its specific implementation process is detailed in FIG. Figure 1 and Figure 2 The embodiment of the label positioning method for the question semantic graph shown is not described in detail here.

[0084] The embodiment further provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above Figure 1 and Figure 2 The label localization method of the question semantic graph shown.

[0085] The embodiment also provides a computer-readable storage medium having a computer program stored thereon, which implements the above Figure 1 and Figure 2 The label localization method of the question semantic graph shown.

[0086] The label positioning method, device, computer equipment and storage medium of the problem semantic graph provided in the above-mentioned embodiments can automatically realize the label positioning of the problem semantic graph. The problem can be easily located without relying on the person locating the problem to master the code debugging ability and problem troubleshooting process, which facilitates the subsequent cross-group docking and problem handling, thereby making the entire process smoother and more efficient.

[0087] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A label positioning method for a question semantic graph, characterized in that: The following steps are involved: Obtain the file to be located reported by the user, wherein the file to be located includes a rendering task ID, a camera ID, and problem data; Obtain a corresponding rendering task result according to the rendering task ID, determine a problem semantic graph from the rendering task result according to the camera ID, and a first mapping relationship between the problem semantic graph and the entity ID; The problem semantic graph is visualized after coloring, the problem pixel is determined from the visualization result combined with the problem data, and the entity ID corresponding to the problem pixel is determined according to the first mapping relationship between the problem semantic graph and the entity ID; The label corresponding to the entity ID is determined from the three-dimensional scene data set according to the entity ID.

2. A label positioning method for a question semantic graph, characterized in that: The following steps are involved: Obtain the file to be located reported by the user, wherein the file to be located includes a rendering task ID, a camera ID, and problem data; Obtain the corresponding rendering intermediate result according to the rendering task ID, determine the problem model graph corresponding to the problem semantic graph from the rendering intermediate result according to the camera ID, and the second mapping relationship between the problem model graph and the entity ID; The problem model graph is visualized after coloring, the problem pixel is determined from the visualization result combined with the problem data, and the entity ID corresponding to the problem pixel is determined according to the second mapping relationship between the problem model graph and the entity ID; The label corresponding to the entity ID is determined from the three-dimensional scene data set according to the entity ID.

3. The label positioning method of the question semantic graph according to claim 1 is characterized in that: The problem data includes problem pixels, and then the entity ID corresponding to the problem pixels is determined according to a first mapping relationship between the problem semantic graph and the entity ID.

4. The label positioning method of the question semantic graph according to claim 2 is characterized in that: The problem data includes problem pixels, and then the entity ID corresponding to the problem pixels is determined according to a second mapping relationship between the problem model graph and the entity ID.

5. The label positioning method of the semantic graph according to claim 1, characterized in that: The obtaining the corresponding rendering task result according to the rendering task ID includes: Get the resource link of the rendering task result according to the rendering task ID, download and parse the json file through the resource link, and get the rendering task result.

6. The label positioning method of the question semantic graph according to claim 2 is characterized in that: The step of obtaining the corresponding rendering intermediate result according to the rendering task ID includes: Get the resource link of the rendering intermediate result according to the rendering task ID, download and parse the json file through the resource link, and get the rendering intermediate result.

7. The label positioning method of the question semantic graph according to claim 1 or 2, characterized in that: Also includes: A question file is generated based on the labels corresponding to the problem pixels in the question semantic graph and supports downloading. The question file includes each pixel value in the question semantic graph and the corresponding label value distribution.

8. A label positioning device for a question semantic graph, characterized in that: include: A file acquisition module, used to acquire the to-be-located file fed back by the user, wherein the to-be-located file includes a rendering task ID, a camera ID, and problem data; A mapping relationship determination module, used to obtain a corresponding rendering task result according to a rendering task ID, determine a problem semantic graph from the rendering task result according to a camera ID, and a first mapping relationship between the problem semantic graph and an entity ID; An entity ID determination module is used to visualize the problem semantic graph after coloring, determine the problem pixel point from the visualization result combined with the problem data, and determine the entity ID corresponding to the problem pixel point according to the first mapping relationship between the problem semantic graph and the entity ID; The label determination module is used to determine the label corresponding to the entity ID from the three-dimensional scene data set according to the entity ID.

9. A label positioning device for a question semantic graph, characterized in that: include: A file acquisition module, which acquires the file to be located reported by the user, wherein the file to be located includes a rendering task ID, a camera ID, and problem data; A mapping relationship determination module, used to obtain a corresponding rendering task result according to a rendering task ID, determine a problem semantic graph from the rendering task result according to a camera ID, and a first mapping relationship between the problem semantic graph and an entity ID; An entity ID determination module is used to visualize the problem model graph after coloring, determine the problem pixel point from the visualization result combined with the problem data, and determine the entity ID corresponding to the problem pixel point according to the second mapping relationship between the problem model graph and the entity ID; The label determination module is used to determine the label corresponding to the entity ID from the three-dimensional scene data set according to the entity ID.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for label positioning of the question semantic graph described in any one of claims 1 to 7 are implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is processed and executed, the steps of the method for label positioning of the question semantic graph described in any one of claims 1-7 are implemented.

Images