CN117036599A - Virtual data set generation method and device, electronic equipment and storage medium - Google Patents

Abstract

The application relates to a virtual data set generation method, a device, electronic equipment and a storage medium, wherein the virtual data set generation method comprises the following steps: loading a three-dimensional model of an object to be identified into a virtual space; shooting the three-dimensional model at a preset shooting point by controlling a virtual camera in the virtual space to obtain an initial virtual data set; and carrying out post-processing on the initial virtual data set to obtain a target virtual data set. Compared with the prior art, the application has the advantages of high efficiency, strong applicability and the like.

Description

A virtual data set generation method, device, electronic equipment and storage medium

Technical field

The present invention relates to the field of data processing technology, and in particular, to a virtual data set generation method, device, electronic equipment and storage medium.

Background technique

Target detection is the most challenging problem in the field of machine vision. Its task is to find the target object in the image and determine the category and location of the target object. It has a wide range of application scenarios and task requirements in industrial production.

At present, target detection algorithms based on artificial neural networks can gradually catch up with traditional detection algorithms in terms of accuracy and versatility. Different from traditional detection algorithms, artificial neural networks not only rely on continuously iterative and refined algorithm models, but also require tens of thousands of high-quality data as model training sets to ensure detection results, which makes the production of data sets a critical step in training detection models. An important part.

At this stage, related technologies still use manual or semi-automatic methods to produce training data sets, resulting in low training efficiency of detection models.

Contents of the invention

The purpose of the present invention is to provide a more efficient virtual data set generation method, device, electronic equipment and storage medium in order to overcome the above-mentioned shortcomings of the prior art.

The object of the present invention can be achieved through the following technical solutions:

According to a first aspect of the present invention, a virtual data set generation method is provided, which method includes:

Load the three-dimensional model of the object to be recognized into the virtual space;

Control a virtual camera in the virtual space to shoot the three-dimensional model at a preset shooting point to obtain an initial virtual data set;

Post-process the initial virtual data set to obtain a target virtual data set.

As a preferred technical solution, the target virtual data set includes: a single target virtual data set;

The single-target virtual data set includes an initial single-target virtual image and a mask image of the initial single-target virtual image;

The post-processing of the initial virtual data set to obtain the target virtual data set includes:

Obtain the rendered contour map of the initial single-target virtual image;

According to the rendered contour map, a mask map of the initial single-target virtual image is obtained.

As a preferred technical solution, the target virtual data set includes: a multi-target virtual data set;

The multi-target virtual data set includes an initial multi-target virtual image and a mask image of the initial multi-target virtual image;

The post-processing of the initial virtual data set to obtain the target virtual data set includes:

Obtain a quasi-solid color rendering of the initial multi-objective virtual image;

According to the quasi-solid color rendering, a mask image of the initial multi-object virtual image is obtained.

As a preferred technical solution, the three-dimensional model includes material information, appearance information and label information of the object to be identified.

As a preferred technical solution, after post-processing the initial virtual data set and obtaining the target virtual data set, the method further includes:

Generate a label file of the target virtual data set according to the label information in the three-dimensional model.

As a preferred technical solution, the preset shooting point is:

A point in the virtual space at a preset distance from the object to be identified;

And/or, a point in the virtual space at a preset longitude;

And/or, a point in the virtual space at a preset latitude.

According to a second aspect of the present invention, an instance segmentation model training method is provided, which method includes:

Using the virtual data set generated by the virtual data set generation method provided by the first aspect or any possible implementation of the first aspect to train the instance segmentation model;

Verify the trained instance segmentation model using a real data set;

The above steps are iterated in a loop until the instance segmentation model meets the verification requirements.

According to a third aspect of the present invention, a virtual data set generating device is provided, which device includes:

The model loading module is used to load the three-dimensional model of the object to be recognized into the virtual space;

An initial virtual data set acquisition module is used to control a virtual camera in the virtual space to shoot the three-dimensional model at a preset shooting point to obtain an initial virtual data set;

A post-processing module is used to perform post-processing on the initial virtual data set to obtain a target virtual data set.

According to a fourth aspect of the present invention, an electronic device is provided, including a memory and a processor. A computer program is stored on the memory. When the processor executes the program, the first aspect or any one of the first aspects is implemented. possible implementation methods.

According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the first aspect or any possible implementation of the first aspect is provided. Methods.

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

1. High efficiency: The virtual data set generation method in the present invention can realize fully automatic generation of virtual data sets, can generate a large number of virtual data sets in a short time for training target detection models, and can save a lot of manpower and time. cost, improve the generation efficiency of virtual data sets, and thus help improve the training efficiency of target detection models.

2. Strong applicability: The virtual data set generation method in the present invention can not only be applied to object detection and recognition scenarios in daily life, but can also be applied to industrial grabbing, object obstacle avoidance and other scenarios where target object outer contour data needs to be obtained. ; In addition, virtual data sets can replace actual scenes, so the virtual data set generation method in the present invention can also be applied in medical surgeries, underwater detection and other scenarios where it is difficult to obtain actual data sets, effectively improving the applicability of the above virtual data set generation method. .

3. A better target detection model can be obtained: When using the virtual data set generated by the virtual data set generation method in the present invention to train a target detection model such as an instance segmentation model, a better target detection effect can be obtained.

Description of the drawings

Figure 1 is a schematic flow chart of a virtual data set generation method in an embodiment of the present invention;

Figure 2 is a model of part 1 used in the production of single-target virtual data sets in the embodiment of the present invention;

Figure 3 is a part model used for making multi-objective virtual data sets in the embodiment of the present invention;

Figure 4 is a schematic flowchart of the acquisition process of the initial virtual data set in the embodiment of the present invention;

Figure 5 is a schematic flow chart of obtaining a mask image of an initial single-target virtual image in an embodiment of the present invention;

Figure 6 is a mixed image of a virtual image and its corresponding mask image in an embodiment of the present invention;

Among them, Figure 6(a) and Figure 6(b) are the single-target mixed images of the initial single-target virtual image and the corresponding mask image of Part 1 and Part 2 respectively, and Figure 6(c) is the single-target mixed image including Figure 6(a) The mixed picture of part 1 in Figure 6(b) and the mixed effect picture of part 2 in Figure 6(b);

Figure 7 is the training loss curve of the instance segmentation model in the embodiment of the present invention;

Figure 8 shows the detection results of real images using the instance segmentation model in the embodiment of the present invention;

Among them, Figure 8(a) shows the detection results of a single target, and Figures 8(b) to 8(e) show the detection results of multiple targets;

Figure 9 is a schematic structural diagram of an electronic device in an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts should fall within the scope of protection of the present invention.

Unless otherwise defined, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the technical field belonging to this application; the terms used herein are for the purpose of describing specific embodiments only and are not intended to be used in Limitation of this application; the terms "including" and "having" and any variations thereof in the description and claims of this application and the above description of the drawings are intended to cover non-exclusive inclusion.

In the description of the embodiments of the present invention, the technical terms "first", "second", etc. are only used to distinguish different objects, and cannot be understood as indicating or implying the relative importance or implicitly indicating the quantity or specificity of the indicated technical features. Sequence or priority relationship. In the description of the embodiments of this application, "plurality" means two or more, unless otherwise explicitly and specifically limited.

Reference in this disclosure to "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment may be included in at least one embodiment of the present application. The appearances of this phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those skilled in the art understand, both explicitly and implicitly, that the embodiments described herein may be combined with other embodiments.

In the description of the embodiment of the present invention, the term "and/or" is only an association relationship describing associated objects, indicating that there can be three relationships, such as A and/or B, which can mean: A exists alone, and A exists simultaneously and B, there are three cases of B alone. In addition, the character "/" in this article generally indicates that the related objects are an "or" relationship.

Figure 1 is a schematic flowchart of a method for generating a virtual data set provided in an embodiment of the present application. This application provides method operation steps as described in the embodiments or flow charts, but more or less operation steps may be included based on conventional or non-inventive efforts. The sequence of steps listed in the embodiment is only one way of executing the sequence of many steps, and does not represent the only execution sequence. This method should be implemented by software and/or hardware. Referring to Figure 1, the method may include:

Step S110: Load the three-dimensional model of the object to be recognized into the virtual space;

Step S120: Control the virtual camera in the virtual space to shoot the three-dimensional model at the preset shooting point to obtain the initial virtual data set;

Step S130: Post-process the initial virtual data set to obtain the target virtual data set.

Optionally, the virtual space in step S110 can be constructed using SolidWorks software, and the three-dimensional model of the object to be identified can be a three-dimensional digital model pre-constructed through SolidWorks software. Of course, the three-dimensional model in the above step S110 can also be constructed by other software.

It can be understood that the three-dimensional model in the above step S110 can be directly constructed by the electronic device that executes the above virtual data set generation method, or can be constructed by other electronic devices and then send the constructed three-dimensional model to the device that executes the above virtual data set generation. Method of electronic equipment.

Optionally, the three-dimensional model in step S110 includes material information, appearance information and label information of the object to be recognized, where the label information can be used to generate a label file of the target virtual data set.

Optionally, in step S110, before loading the three-dimensional model of the object to be recognized into the virtual space, a virtual background of the virtual space may be set, where the virtual background includes background environment, object position, lighting parameters, etc.

It can be understood that the three-dimensional model is also divided into a single-target three-dimensional model and a multi-target three-dimensional model. The single-target three-dimensional model only includes one target to be identified, and the multi-target three-dimensional model includes multiple targets to be identified. The single-objective 3D model can be a part model as shown in Figure 2, and the multi-objective 3D model can be an assembly model as shown in Figure 3. The relative positions between multiple objects can be determined by the relative positions between parts in the assembly model. to make sure.

Optionally, step S120 can use the built-in rendering plug-in photoview360 in SolidWorks to perform rendering and shooting. For the rendering parameter settings during rendering and shooting, refer to Table 1. The contour color of the rendered contour map in the single-target virtual data set can be set according to the virtual background color and the three-dimensional model color. It is necessary to ensure that the contour can be well distinguished from the virtual background and the three-dimensional model.

Table 1 Rendering shooting parameters

Optionally, the preset shooting point in step S120 is: a point in the virtual space at a preset distance from the object to be identified; and/or a point in the virtual space at a preset longitude; and/or a point in the virtual space point at a preset latitude.

It can be understood that this embodiment can also modify the longitude and latitude of the camera in space and the distance from the object to achieve all-round continuous shooting of the target object. The specific implementation method is shown in Figure 4. Among them, L (unit: mm) is the distance from the camera to the target object; A1 (unit: °) is the longitude value in the spherical coordinate system where the camera is located; A2 (unit: °) is the latitude in the spherical coordinate system where the camera is located value; q is a counter that counts the number of virtual images generated by rendering and provides a serial number for saving as a file name. It can be understood that the implementation shown in Figure 4 is also equivalent to controlling a virtual camera to shoot at a preset shooting point. The changing process of the longitude and latitude values and the distance value between the camera and the object to be identified is determined by a preset shooting point. The process of moving to another preset shooting point.

Optionally, the target virtual data set includes: a single-target virtual data set; the single-target virtual data set includes an initial single-target virtual image and a mask map of the initial single-target virtual image;

At this time, step S130 performs post-processing on the initial virtual data set to obtain the target virtual data set, including: obtaining a rendering outline of the initial single-target virtual image; and obtaining a mask of the initial single-target virtual image based on the rendering outline. Code map.

It can be understood that the above-mentioned initial single-target virtual image and rendered contour map can be obtained by rendering twice in the same camera perspective, and saved in JPG format and BMP format respectively. In SolidWorks software, the rendering parameters ContourLineThickness and ContourLineColor can be set so that the rendered image contains surface feature contours, which facilitates distinguishing the target object from the background and facilitates subsequent data processing.

Optionally, the mask map of the initial single-target virtual image can be an index map file with a bit depth of 8, and the acquisition method can be: as shown in Figure 5, after obtaining the rendering contour map of the initial single-target virtual image, render The contour map contains colored contours (such as red contours). The three RGB channels can be traversed according to the color category to extract the contours from the rendered contour map; then the image is binarized and the target pixel is judged by Connectivity fills holes inside the outline; finally, the image is converted into the required format. On the basis of retaining the original data, each pixel in the image is converted from binary to 8-bit depth, and then installed in the index table and the image is rewritten. For the index image format, the image format is finally modified from BMP to PNG format to obtain the mask image of the initial single-target virtual image.

Optionally, the target virtual data set includes: a multi-target virtual data set; the multi-target virtual data set includes an initial multi-target virtual image and a mask map of the initial multi-target virtual image;

At this time, step S130 performs post-processing on the initial virtual data set to obtain the target virtual data set, including: obtaining a quasi-solid color rendering of the initial multi-target virtual image; and obtaining a mask image of the initial multi-target virtual image based on the quasi-solid color rendering. .

It can be understood that multi-target data contains multiple target objects in one picture, and these objects need to be associated with their respective labels. This embodiment uses a solid-color rendering method to provide data support for subsequent generation of mask files. The rendering method of the initial multi-target virtual image is to process the same set of parts in batches. One set is used for rendering to generate the original image, and the other set is used for solid-color rendering. The two sets maintain the same placement of parts and ensure that the camera is rotated each time. The angle is exactly the same as the changed distance, thus matching the original image and the mask to shorten the running time of the algorithm.

According to the quasi-solid color rendering, the method of obtaining the mask map of the initial multi-target virtual image can be:

The initial multi-target virtual image identifies the colors corresponding to each target part according to the color order in the index table. Combined with the generated label information, it traverses all pixels in the image and determines them to determine which part the pixel belongs to. , if it does not belong to any of them, it will be judged as background;

The coordinates of the determined pixels are stored in each matrix respectively. After binary operation, hole filling and format conversion of each matrix, independent masks of each target can be obtained in three different matrices. This At this time, the background pixel value of each independent mask is 0, and the mask pixel value is 1. The pixel values of the target areas in each color matrix are modified according to the order of the index table and then merged;

After completing the production of the data matrix, load the index table and modify it to save it in PNG format.

It can be understood that this embodiment designs an overlap detection algorithm, and performs a data superposition before modifying each matrix data. At this time, the mask pixel values are all 1, so the part with a pixel value greater than 1 after superposition is the overlapped part. . Overlap only occurs at the transition between the edges of two or more parts. Set the pixel value of the overlapping part to zero, making it a black background.

Optionally, after obtaining the mask map of the initial multi-target virtual image for each target based on the solid-color rendering, you can create a mixed overlay image by calling the imfuse function in MATLAB, and use alpha blending to cover the virtual original image and mask. image, jointly scale the intensity values in the image so that they are mixed into the same image. The mixed image of the original image and the mask image is shown in Figure 6, where Figure 6(a) and Figure 6(b) are part 1 respectively. and the single-target mixed image of the initial single-target virtual image of Part 2 and the corresponding mask map. Figure 6(c) is a mixed image containing the mixed image of Part 1 in Figure 6(a) and the mixed image of Part 2 in Figure 6(b) renderings;

Optionally, in step S130, after post-processing the initial virtual data set and obtaining the target virtual data set, it also includes: generating a label file of the target virtual data set according to the label information in the three-dimensional model.

In this embodiment, the part label is directly written into the Excel file, and the source of the label is the preset part name in the three-dimensional model.

The part names of multi-object data belong to the assembly, and similar parts may exist. In the case where there are multiple and uncertain part names in the assembly, the number and name information of the visible parts in the current camera perspective is read through the API interface. According to the number of visible parts, the name tags are read in sequence and sequenced. Write to Excel. When there are multiple parts of the same type, different parts of the same type are distinguished by suffixes. You can load the PatternRecognition Toolbox additional function resource in MATLAB, and MATLAB reads the content of the Excel intermediate file containing label information, and finally saves it to a YML file.

The above is an introduction to the embodiments of the virtual data set generation method in the present invention. The following is an introduction to the embodiments of the instance segmentation model training method to further explain the solution in the present invention.

An embodiment of the present invention also provides an instance segmentation model training method, which method includes:

Use the virtual data set generated by the above virtual data set generation method to train the instance segmentation model;

Use real data sets to verify the trained instance segmentation model;

The above steps are iterated in a loop until the instance segmentation model meets the verification requirements.

The above instance segmentation model can use the Mask R-CNN model. The virtual data set consists of 504 virtual images of part 1, 504 virtual images of part 2, and 504 multi-target virtual images containing part 1 and part 2.

This embodiment builds a GPU-based deep learning computing platform, and the hardware parameters of the platform are shown in Table 2. The software runs on Windows 10 system, the deep learning framework uses TensorFlow and Keras, and the programming language uses Python 3.6. Configure Python's data processing library and image processing library, such as OpenCV, Numpy, Pillow, Scikit, etc.

Table 2 Hardware configuration of computing platform

Use the above virtual data set to train the Mask R-CNN detection model. The network architecture is ResNet-101, COCO's pre-training weights are used, the learning rate is 0.001, and a total of 100 epochs are trained. The loss curve of model training is shown in Figure 7.

Use the trained instance segmentation model to detect real images containing Part 1 and Part 2. The detection results are shown in Figure 8. By analyzing the detection results, it can be concluded that the detection model trained by the virtual data set can better instance segment the target objects in the real environment. Figure 8(a) shows that the model has good segmentation capabilities for single-target real images. The backgrounds in Figures 8(b) to 8(e) are dotted patterns that have never appeared in the data set, and there are varying degrees of background reflection. In Figure 8(d), a cast iron part with a slightly different shape is detected. 2. This material does not appear in the virtual data set provided for training. The detection results show that the detection model has good generalization ability and can identify parts of different materials or slightly different shapes. 2. The non-target parts in Figure 8(e) were not detected, indicating that the detection model has a good ability to eliminate interference items.

This embodiment also performs a statistical analysis on the instance segmentation accuracy of the two types of parts. The instance segmentation accuracy and classification accuracy of the two types of targets are shown in Table 3. Among them, the instance segmentation accuracy IoU refers to the ratio of the intersection of the segmentation detection result and the actual result of the object Ground Truth and its union. It is a dimensionless number. The theoretical maximum value is 1. The closer to 1, the higher the segmentation accuracy; the accuracy precision is It refers to the proportion of samples whose predicted value is true that is actually true; recall refers to the proportion of all samples that are actually true that are detected. If a single index is high and the other index is too low, it indicates that the detection model has a classification problem.

Table 3 Instance segmentation accuracy and classification accuracy of two types of targets

target Segmentation accuracy IoU(%) Accuracy precision(%) Recall rate recall(%) Part 1 82.4 97.6 96.4 Part 2 85.5 98.8 96.5

The analysis results show that the virtual data set produced using this case can converge during the Mask R-CNN network model training process, and at the same time has good detection results in the detection of real pictures, verifying the feasibility of the virtual data set production method .

The above is an introduction to the method embodiments. The solution of the present invention will be further described below through device embodiments.

An embodiment of the present invention also provides a virtual data set generating device, which includes:

The model loading module is used to load the three-dimensional model of the object to be recognized into the virtual space;

An initial virtual data set acquisition module is used to control a virtual camera in the virtual space to shoot the three-dimensional model at a preset shooting point to obtain an initial virtual data set;

A post-processing module is used to perform post-processing on the initial virtual data set to obtain a target virtual data set.

Optionally, the target virtual data set obtained by the post-processing module includes: a single target virtual data set;

The single-target virtual data set includes an initial single-target virtual image and a mask image of the initial single-target virtual image;

At this time, the post-processing module is specifically configured to: obtain a rendering contour map of the initial single-target virtual image; and obtain a mask map of the initial single-target virtual image based on the rendering contour map.

Optionally, the target virtual data set acquired by the post-processing module includes: a multi-target virtual data set; the multi-target virtual data set includes an initial multi-target virtual image and a mask image of the initial multi-target virtual image;

At this time, the post-processing module is specifically configured to: obtain a quasi-solid color rendering of the initial multi-objective virtual image; and obtain a mask image of the initial multi-objective virtual image based on the quasi-pure color rendering.

Optionally, the three-dimensional model loaded into the virtual space by the model loading module includes the material information, appearance information and label information of the object to be identified.

Optionally, the virtual data set generation device also includes:

A label file generation module, configured to generate a label file for the target virtual data set according to the label information in the three-dimensional model.

Optionally, the preset shooting point in the initial virtual data set acquisition module is: a point in the virtual space at a preset distance from the object to be identified; and/or a point in the virtual space at a preset longitude. point; and/or, a point at a preset latitude in the virtual space.

Figure 9 shows a schematic block diagram of an electronic device that may be used to implement embodiments of the present disclosure. As shown in Figure 9, the electronic device of the present invention includes a central processing unit (CPU), which can be loaded from a storage unit into a random access memory (RAM) according to computer program instructions stored in a read-only memory (ROM). instructions to perform various appropriate actions and processing. In RAM, various programs and data required for device operation can also be stored. CPU, ROM and RAM are connected to each other through buses. Input/output (I/O) interfaces are also connected to the bus.

Multiple components in the device are connected to the I/O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of monitors, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, For example, network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information/data with other devices through computer networks such as the Internet and/or various telecommunications networks.

The processing unit executes each of the methods and processes described above, such as steps S110 to S130 of the method of the present invention. For example, in some embodiments, steps S110 to S130 of the method of the present invention can be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and/or installed onto the device via the ROM and/or communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more of the above-described steps S110 to S130 of the method of the present invention may be executed. Alternatively, in other embodiments, the CPU may be configured to execute steps S110 to S130 of the method of the present invention in any other suitable manner (for example, by means of firmware).

The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), Application Specific Standard Products (ASSPs), Systems on Chips (SOCs), Complex Programmable Logical device (CPLD) and so on.

Program code for implementing the methods of the invention may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that the program codes, when executed by the processor or controller, cause the functions specified in the flowcharts and/or block diagrams/ The operation is implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media would include one or more wire-based electrical connections, laptop disks, hard drives, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.

The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technical field can easily think of various equivalent methods within the technical scope disclosed in the present invention. Modifications or substitutions shall be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims (10)

1. A method for generating a virtual data set, characterized in that the method includes: Load the three-dimensional model of the object to be recognized into the virtual space; Control a virtual camera in the virtual space to shoot the three-dimensional model at a preset shooting point to obtain an initial virtual data set; Post-process the initial virtual data set to obtain a target virtual data set. 2. A method for generating a virtual data set according to claim 1, wherein the target virtual data set includes: a single target virtual data set; The single-target virtual data set includes an initial single-target virtual image and a mask image of the initial single-target virtual image; The post-processing of the initial virtual data set to obtain the target virtual data set includes: Obtain the rendered contour map of the initial single-target virtual image; According to the rendered contour map, a mask map of the initial single-target virtual image is obtained. 3. A method for generating a virtual data set according to claim 1, characterized in that the target virtual data set includes: a multi-target virtual data set; The multi-target virtual data set includes an initial multi-target virtual image and a mask image of the initial multi-target virtual image; The post-processing of the initial virtual data set to obtain the target virtual data set includes: Obtain a quasi-solid color rendering of the initial multi-objective virtual image; According to the quasi-solid color rendering, a mask image of the initial multi-object virtual image is obtained. 4. A method for generating a virtual data set according to any one of claims 1 to 3, characterized in that the three-dimensional model includes material information, appearance information and label information of the object to be identified. 5. A method for generating a virtual data set according to claim 4, characterized in that, after post-processing the initial virtual data set and obtaining the target virtual data set, it further includes: Generate a label file of the target virtual data set according to the label information in the three-dimensional model. 6. A method for generating a virtual data set according to any one of claims 1 to 3, characterized in that the preset shooting point is: A point in the virtual space at a preset distance from the object to be identified; And/or, a point in the virtual space at a preset longitude; And/or, a point in the virtual space at a preset latitude. 7. An instance segmentation model training method, characterized in that the method includes: Using the virtual data set generated by the virtual data set generation method according to any one of claims 1 to 6 to train the instance segmentation model; Verify the trained instance segmentation model using a real data set; The above steps are iterated in a loop until the instance segmentation model meets the verification requirements. 8. A virtual data set generating device, characterized in that the device includes: The model loading module is used to load the three-dimensional model of the object to be recognized into the virtual space; An initial virtual data set acquisition module is used to control a virtual camera in the virtual space to shoot the three-dimensional model at a preset shooting point to obtain an initial virtual data set; A post-processing module is used to perform post-processing on the initial virtual data set to obtain a target virtual data set. 9. An electronic device, including a memory and a processor, with a computer program stored on the memory, characterized in that when the processor executes the program, the implementation as described in any one of claims 1 to 6 or 7 is achieved. Methods. 10. A computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and the computer program is executed by a processor as described in any one of claims 1 to 6 or 7. method.