Automatic labeling method, computer device and storage medium for segmentation tasks

Abstract

The present application discloses an automatic labeling method, computer equipment and storage medium for segmentation tasks, and relates to the field of image processing technology. The method of the present application includes: obtaining real data to be labeled corresponding to the segmentation task; obtaining a segmentation labeling model based on part of the real data to be labeled and a pre-trained over-segmentation prediction model, and automatically labeling the real data to be labeled corresponding to the segmentation task based on the segmentation labeling model. The segmentation labeling model adopted in the present application inherits the high-precision performance of the over-segmentation prediction model, which can be adapted to different segmentation tasks and has good versatility. Automatic labeling of the real data to be labeled based on the segmentation labeling model with high-precision performance and good versatility of the present application can greatly improve the efficiency of automatic labeling.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an automatic labeling method, computer equipment and storage medium for segmentation tasks. Background Art

[0002] With the development of artificial intelligence in recent years, the demand for image segmentation has increased, and various types of image segmentation models have emerged. When processing image segmentation tasks based on image segmentation models, the segmented targets can be labeled at the same time. For example, based on existing image segmentation models, semantic segmentation, instance segmentation, panoramic segmentation, over-segmentation and other segmentation tasks can be processed.

[0003] At present, some of the labeling methods used for segmentation tasks are based on the trained segmentation model for offline segmentation labeling, and then modify the labeling results according to the task requirements. This method requires the segmentation model to be trained in advance. However, since many labeling tasks have special customization requirements, it is necessary to train a new model according to the actual task each time. Although this method has high accuracy, it has poor universality. Others are based on online tools for labeling, which do not require a universal model, but each labeling requires manual clicks. The cost of this method is that each labeling requires a lot of manual intervention and adjustment, and the accuracy is not high. Summary of the invention

[0004] In order to overcome the above defects, the present invention provides an automatic labeling method, computer device and storage medium for segmentation tasks, which can solve the problems of poor versatility or high labor cost in the segmentation labeling in the prior art.

[0005] In a first aspect, the present invention provides an automatic annotation method for a segmentation task, the method comprising:

[0006] Obtain the real data to be labeled corresponding to the segmentation task;

[0007] Obtaining a segmentation and annotation model based on part of the real data to be annotated and a pre-trained over-segmentation prediction model;

[0008] Based on the segmentation and labeling model, the real data to be labeled corresponding to the segmentation task is automatically labeled.

[0009] In some embodiments, the method further includes: obtaining the over-segmentation prediction model based on training of over-segmentation synthetic data.

[0010] Furthermore, the over-segmentation prediction model obtained by training based on over-segmentation synthetic data includes:

[0011] Obtain historical real image data;

[0012] Obtaining segmented annotation data based on the historical real image data and a preset image segmentation model;

[0013] Performing generalization processing based on the over-segmented labeled data to obtain the over-segmented synthetic data;

[0014] The over-segmentation prediction model is obtained based on the over-segmentation synthetic data training.

[0015] In some embodiments, obtaining a segmentation annotation model based on part of the real data to be annotated and a pre-trained over-segmentation prediction model includes:

[0016] Performing transfer learning on the over-segmentation prediction model based on part of the real data to be labeled to obtain the segmentation labeling model;

[0017] Furthermore, the performing transfer learning on the over-segmentation prediction model based on part of the real data to be labeled to obtain the segmentation labeling model includes:

[0018] Obtaining an over-segmentation prediction result based on part of the real data to be labeled and the over-segmentation prediction model;

[0019] Merging the image segmentation blocks corresponding to the over-segmentation prediction results to obtain a segmentation annotation result corresponding to the segmentation task;

[0020] The segmentation and annotation model is obtained by performing transfer learning on the over-segmentation prediction model based on part of the real data to be annotated and the segmentation and annotation results.

[0021] Furthermore, performing transfer learning on the over-segmentation prediction model based on part of the real data to be labeled and the segmentation labeling result to obtain the segmentation labeling model includes:

[0022] The pre-trained over-segmentation prediction model is used as a teacher model, and the teacher model is distilled based on part of the real data to be labeled to obtain the segmentation and labeling model.

[0023] Furthermore, obtaining the over-segmentation prediction result based on part of the real data to be labeled and a pre-trained over-segmentation prediction model includes: inputting part of the real data to be labeled into the pre-trained over-segmentation prediction model, and obtaining the over-segmentation prediction result according to the model output, wherein the data amount of the part of the real data to be labeled is less than a preset threshold.

[0024] Furthermore, merging the image segmentation blocks corresponding to the over-segmentation prediction results to obtain the segmentation annotation results corresponding to the segmentation task includes:

[0025] Determining a segmentation target according to the segmentation task;

[0026] The image segmentation blocks corresponding to the over-segmentation prediction results are merged according to the segmentation target to obtain a segmentation target area, and a target label is assigned to the segmentation target area to obtain the segmentation annotation result, wherein the target label includes a semantic label and / or an instance label.

[0027] Furthermore, merging the image segmentation blocks corresponding to the over-segmentation prediction results according to the segmentation target to obtain the segmentation target area includes: determining semantic features of the segmentation target, and merging adjacent image segmentation blocks according to the semantic features to obtain the segmentation target area.

[0028] Furthermore, the training of obtaining the segmentation labeling model based on the real data to be labeled, the segmentation labeling result and the over-segmentation prediction model includes: based on part of the real data to be labeled and the segmentation labeling result, performing model distillation on the over-segmentation prediction model to obtain the segmentation labeling model.

[0029] In a second aspect, the present invention provides a computer device comprising a processor and a memory, wherein the memory is suitable for storing multiple program codes, and the program codes are suitable for being loaded and run by the processor to execute the automatic labeling method for segmentation tasks described in any one of the technical solutions of the above-mentioned automatic labeling method for segmentation tasks.

[0030] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of program codes, wherein the program codes are suitable for being loaded and run by a processor to execute the automatic labeling method for segmentation tasks described in any one of the technical solutions of the above-mentioned automatic labeling method for segmentation tasks.

[0031] The above one or more technical solutions of the present invention have at least one or more of the following beneficial effects: when processing the real data to be labeled corresponding to the segmentation task, the present invention obtains a segmentation labeling model based on part of the real data to be labeled and a pre-trained over-segmentation prediction model, and automatically labels the real data to be labeled corresponding to the segmentation task based on the segmentation labeling model. The segmentation labeling model obtained based on part of the real data to be labeled and the pre-trained over-segmentation model in the present invention inherits the high-precision performance of the over-segmentation prediction model, and can be adapted to different segmentation tasks with good versatility. Furthermore, the present invention uses the segmentation synthetic data obtained by generalization processing as training data when training the over-segmentation prediction model, thereby improving the generalization of the over-segmentation prediction model, and correspondingly improving the generalization of the segmentation labeling model obtained through transfer learning, so that the processing of segmentation tasks based on the segmentation labeling model of the present invention can greatly improve the efficiency of automatic labeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The disclosure of the present invention will become more easily understood with reference to the accompanying drawings. It is easy for those skilled in the art to understand that these drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. In addition, similar numbers in the figures are used to represent similar components, among which:

[0033] Figure 1 is a schematic flow chart of main steps of an automatic annotation method for segmentation tasks according to an embodiment of the present application;

[0034] Figure 2 This is a flowchart of a specific implementation method of obtaining an over-segmentation prediction model based on over-segmentation synthetic data training provided by the present application;

[0035] Figure 3 yes Figure 1 A specific implementation method flow of step S12 shown in FIG.

[0036] Figure 4 is a schematic block diagram of an automatic annotation system for segmentation tasks according to an embodiment of the present application;

[0037] Figure 5 It is a main structural diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0038] Some embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0039] In the description of the present invention, "module" and "processor" may include hardware, software or a combination of the two. A module may include hardware circuits, various suitable sensors, communication ports, and memories, and may also include software parts, such as program codes, or a combination of software and hardware. The processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. The processor has data and / or signal processing functions. The processor may be implemented in software, hardware, or a combination of the two. Non-temporary computer-readable storage media include any suitable medium that can store program codes, such as a magnetic disk, a hard disk, an optical disk, a flash memory, a read-only memory, a random access memory, and the like. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The term "at least one A or B" or "at least one of A and B" has a similar meaning to "A and / or B", and may include only A, only B, or A and B. The singular terms "one" and "the" may also include plural forms.

[0040] The technical terms involved in the embodiments of this application are explained as follows:

[0041] Semantic segmentation: is the process of dividing an image into multiple regions, each of which is associated with a specific category, such as a person, dog, or vehicle. The result is that each pixel in the image is labeled as belonging to a certain category. Semantic segmentation focuses on "what is this pixel". ‌

[0042] Instance segmentation: not only segment the image into regions associated with a specific category, but also distinguish different individuals in the same category. For example, if there are two cars in the image, semantic segmentation will label them both as "car", while instance segmentation will label them as "car 1" and "car 2" to distinguish different instances. Instance segmentation focuses not only on "what is this pixel", but also on "which one it is". ‌

[0043] Panoramic segmentation: Combining the characteristics of semantic segmentation and instance segmentation, it aims to process all pixels in the image at the same time and assign a semantic label and an instance ID to each pixel. This segmentation method not only distinguishes different objects (instances) in the image, but also classifies continuous areas such as the background (semantics), thereby generating a global, unified segmented image.

[0044] Over-segmentation: usually refers to segmenting every local area in the image into a separate object, even if these areas are not visually distinct.

[0045] SAM model: is a large visual model. SAM performs cueable segmentation, which is different from semantic segmentation in two aspects: (1) the masks generated by SAM have no labels; (2) SAM relies on cue words (such as point, box, mask, etc.).

[0046] Transfer learning: is a machine learning method whose core concept is to transfer the knowledge or model learned in one task (source task) to another related task (target task) to improve the performance of the new task. It can also be understood as taking the model developed for task A as the starting point and reusing it in the process of developing the model for task B.

[0047] For image segmentation tasks, semantic segmentation tasks require the pixel categories in the input image to be segmented, while instance segmentation requires not only the pixel categories but also the distinction between different target individuals. Taking human body segmentation as an example, semantic segmentation tasks only require the segmentation of which pixels in the image are / are not human bodies, while instance segmentation requires the algorithm to distinguish different human individuals to which each pixel belongs, in addition to the requirements of semantic segmentation. Annotation for segmentation tasks has always been a time-consuming and costly task in image annotation tasks. The main reason is that segmentation tasks often require high precision and require the classification of each pixel, which is time-consuming for small objects and complex boundaries and details. Therefore, automatic annotation is usually used to improve the efficiency of annotation and reduce the content that needs to be annotated.

[0048] See attached Figure 1 , Figure 1 This is a flowchart of the main implementation steps of an automatic annotation method for segmentation tasks provided by an embodiment of the present application. As shown in the figure, the method mainly includes the following steps S11 to S13:

[0049] Step S11: Obtain the real data to be labeled corresponding to the segmentation task;

[0050] In this embodiment, the real data to be annotated can be understood as the real image data used to perform the segmentation task, wherein performing the segmentation task can be understood as a process of segmenting the image data to be segmented according to the specified segmentation target to obtain annotated image segments, wherein the annotations of the image segments can be semantic labels and / or instance labels corresponding to the segmentation target. In combination with practical applications, it can be known that the real data to be annotated corresponding to the segmentation task obtained in this step is usually in large quantities.

[0051] Step S12: obtaining a segmentation and annotation model based on part of the real data to be annotated and a pre-trained over-segmentation prediction model;

[0052] In this embodiment, the over-segmentation prediction model can be a model for image over-segmentation obtained by training an existing convolutional neural network model (such as a Unet model, a SAM model, a PSPNet model, etc.). It should be understood that the training data label uses an over-segmentation label, and the loss function used in model training is usually a binary cross entropy loss function (BCE Loss). The BCE Loss formula is as follows:

[0053]

[0054] where y is a binary label 0 or 1, is the probability that the output belongs to the label, and N represents the number of groups of objects predicted by the model.

[0055] Specifically in this embodiment, this step can be specifically to obtain the segmentation and annotation model by performing transfer learning on the over-segmentation prediction model based on part of the real data to be annotated. In practical applications, the image data obtained by processing the output of the over-segmentation prediction model according to the segmentation task can be used as training data for the segmentation and annotation model. It should be understood that the output of the over-segmentation prediction model is an over-segmentation prediction result. For example, the over-segmentation prediction result includes segmentation block 1, segmentation block 2, and segmentation block 3 belonging to object A. The segmentation task is to segment object A from the image. Then, the image data obtained by processing the output of the segmentation prediction model according to the segmentation task can be understood as image data with object A annotated. The segmentation and annotation model obtained by training the image data as training data can be used to process the segmentation task of annotating object A.

[0056] Step S13: Automatically label the real data to be labeled corresponding to the segmentation task based on the segmentation labeling model.

[0057] In a specific implementation, the over-segmentation prediction model in the above step S12 can be obtained by training based on over-segmentation synthetic data. Training the over-segmentation prediction model with synthetic data can improve the generalization of the over-segmentation prediction model, thereby making the subsequent segmentation annotation model based on the over-segmentation prediction model more accurate in its annotation effect.

[0058] like Figure 2 FIG. 1 is a flowchart of a specific implementation method for obtaining an over-segmentation prediction model based on over-segmentation synthetic data training provided by the present embodiment, which mainly includes the following steps S111 to S114:

[0059] Step S111: Acquire historical real image data;

[0060] Step S112: obtaining segmented annotation data based on the historical real image data and a preset image segmentation model;

[0061] Specifically in this embodiment, the preset image segmentation model can adopt an existing trained SAM model, input historical real image data into the SAM model, and determine the over-segmented annotation data according to the mask image output by the model.

[0062] Step S113: performing generalization processing based on the over-segmented labeled data to obtain the over-segmented synthetic data;

[0063] Specifically, in this embodiment, an existing image generation model (such as a Diffusion model) may be used, and the over-segmentation annotation data may be used as a control condition (control) to guide the generation of the over-segmentation synthetic data.

[0064] Step S114: obtaining the over-segmentation prediction model based on the over-segmentation synthetic data training.

[0065] In this embodiment, this step may be to use the over-segmentation synthetic data as training data to train an existing convolutional neural network model (such as a U-Net model) to obtain the over-segmentation prediction model.

[0066] Based on the above step S12, the embodiment of the present application provides a specific implementation method for obtaining a segmentation annotation model based on part of the real data to be annotated and a pre-trained over-segmentation prediction model, such as Figure 3 As shown, it mainly includes the following steps S121 to S123:

[0067] Step S121: obtaining an over-segmentation prediction result based on part of the real data to be labeled and a pre-trained over-segmentation prediction model;

[0068] In this embodiment, this step can be specifically as follows: inputting part of the real data to be labeled into the pre-trained over-segmentation prediction model, and obtaining the over-segmentation prediction result according to the model output, wherein the data amount of the part of the data is less than a preset threshold. In practical applications, the preset threshold can be custom set. Generally, the data amount corresponding to the preset threshold refers to a small amount of data relative to the total amount of data.

[0069] Step S122: merging the image segmentation blocks corresponding to the over-segmentation prediction results to obtain a segmentation annotation result corresponding to the segmentation task;

[0070] In this embodiment, this step may be specifically as follows: determining a segmentation target according to the segmentation task; merging the image segmentation blocks corresponding to the over-segmentation prediction results according to the segmentation target to obtain a segmentation target area, and assigning a target label to the segmentation target area to obtain the segmentation annotation result, wherein the target label includes a semantic label and / or an instance label. Semantic labels are used to distinguish different object categories, such as classifying all "cats" in an image into one category without distinguishing specific individual cats, and instance labels not only distinguish the category of the object, but also distinguish different individuals. For example, among the multiple dogs detected, each dog is labeled as a different instance. ‌‌

[0071] In a specific implementation, the merging of the image segments corresponding to the over-segmentation prediction results according to the segmentation target to obtain the segmentation target region can be specifically as follows: determining the semantic features of the segmentation target, and merging adjacent image segments according to the semantic features to obtain the segmentation target region. Among them, the semantic features can be understood as image features corresponding to semantic categories, and the determination of adjacent image segments can be to determine adjacent image segments using the connectivity of the segmentation line or to determine adjacent image segments by edge detection. Exemplarily, in this embodiment, a two-stage target detection network can be used to realize merging the image segments according to the segmentation target to obtain the segmentation target region. It should be understood that the two-stage target detection network can be used to detect whether the image segments belong to the same segmentation target and to predict whether the segmentation lines of the image segments are connected. In practical applications, adjacent segments that belong to the same segmentation target and whose segmentation lines are connected can be merged based on the two-stage target detection network.

[0072] Step S123: performing transfer learning on the over-segmentation prediction model based on part of the real data to be labeled and the segmentation labeling result to obtain the segmentation labeling model.

[0073] In this embodiment, part of the real data to be labeled can be used as training data, and the segmentation labeling results can be used as labels. The over-segmentation prediction model can be trained through transfer learning to obtain the segmentation labeling model. The objective function used in model training can also use BCE Loss. It should be understood that the segmentation labeling model obtained through transfer learning has learned the knowledge of the over-segmentation prediction model and inherited the performance of the over-segmentation prediction model. Furthermore, the segmentation labeling model can also be a lightweight model obtained on the basis of the over-segmentation prediction model.

[0074] Exemplarily, this step may be specifically as follows: using the pre-trained over-segmentation prediction model as a teacher model, and performing model distillation on the teacher model based on part of the real data to be labeled to obtain the segmentation and labeling model.

[0075] Furthermore, using the pre-trained over-segmentation prediction model as a teacher model, and performing model distillation on the teacher model based on part of the real data to be labeled to obtain the segmentation and labeling model can be specifically as follows: using the over-segmentation prediction model as a teacher model, and training a pre-designed student model based on part of the real data to be labeled and the segmentation and labeling results output by the teacher model to obtain the segmentation and labeling model.

[0076] See attached Figure 4The present application also provides an automatic annotation system for segmentation tasks. The system shown in the figure mainly includes a data acquisition module 201, a model training module 202 and an automatic annotation module 203, wherein:

[0077] The data acquisition module 201 is used to acquire the real data to be labeled corresponding to the segmentation task;

[0078] The model training module 202 is used to obtain a segmentation and annotation model based on part of the real data to be annotated and a pre-trained over-segmentation prediction model.

[0079] The segmentation and annotation model learns the knowledge of the over-segmentation prediction model and inherits the performance of the over-segmentation prediction model.

[0080] The automatic labeling module 203 is used to automatically label the real data to be labeled corresponding to the segmentation task based on the segmentation labeling model.

[0081] It should be understood that since the setting of each module is only for illustrating the functional units of the system of the present invention, the physical devices corresponding to these modules may be the processor itself, or a part of the software in the processor, a part of the hardware, or a part of the combination of software and hardware. Therefore, the number of each module in the figure is only schematic.

[0082] Those skilled in the art will appreciate that the modules in the system may be adaptively split or merged. Such splitting or merging of specific modules will not cause the technical solution to deviate from the principle of the present invention, and therefore, the technical solutions after splitting or merging will fall within the protection scope of the present invention.

[0083] It is understood by those skilled in the art that the present invention implements all or part of the processes in the method of the above embodiment, and can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium, etc. that can carry the computer program code.

[0084] Furthermore, the present invention also provides a computer device. Figure 5 , Figure 5It is a schematic diagram of the main structure of a computer device according to an embodiment of the present application. In a computer device embodiment according to the present application, the computer device includes a processor 301 and a memory 302. The memory 302 can be configured to store a program for executing the automatic labeling method for segmenting tasks of the above method embodiment. The processor 301 can be configured to execute the program in the memory, which includes but is not limited to the program for executing the automatic labeling method for segmenting tasks of the above method embodiment. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer device can be a control device device formed by various electronic devices.

[0085] In some possible implementations of the present application, the computer device may include multiple processors 301 and multiple memories 302. The program for executing the automatic annotation method for segmenting tasks of the above method embodiment can be divided into multiple subprograms, and each subprogram can be loaded and run by the processor 301 to execute different steps of the automatic annotation method for segmenting tasks of the above method embodiment. Specifically, each subprogram can be stored in different memories 302, and each processor 301 can be configured to execute the program in one or more memories 302 to jointly implement the automatic annotation method for segmenting tasks of the above method embodiment, that is, each processor 301 executes different steps of the automatic annotation method for segmenting tasks of the above method embodiment, to jointly implement the automatic annotation method for segmenting tasks of the above method embodiment.

[0086] Furthermore, the present invention also provides a computer-readable storage medium. In a computer-readable storage medium embodiment according to the present invention, the computer-readable storage medium can be configured to store a program for executing the automatic labeling method for segmentation tasks of the above-mentioned method embodiment, and the program can be loaded and run by the processor to implement the above-mentioned automatic labeling method for segmentation tasks. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium can be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present invention is a non-temporary computer-readable storage medium.

[0087] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. An automatic labeling method for segmentation tasks, characterized in that: The method comprises: Obtain the real data to be labeled corresponding to the segmentation task; A segmentation and annotation model is obtained based on part of the real data to be annotated and a pre-trained over-segmentation prediction model; the over-segmentation prediction model is a model for image over-segmentation obtained by training a convolutional neural network, and a training data label of the over-segmentation prediction model uses an over-segmentation label; Automatically labeling the real data to be labeled corresponding to the segmentation task based on the segmentation labeling model; The obtaining of the segmentation annotation model based on part of the real data to be annotated and the pre-trained over-segmentation prediction model comprises: Obtaining an over-segmentation prediction result based on part of the real data to be labeled and the over-segmentation prediction model; Merging the image segmentation blocks corresponding to the over-segmentation prediction results to obtain a segmentation annotation result corresponding to the segmentation task; The segmentation and annotation model is obtained by performing transfer learning on the over-segmentation prediction model based on part of the real data to be annotated and the segmentation and annotation results.

2. The method according to claim 1, characterized in that The method further includes: obtaining the over-segmentation prediction model based on over-segmentation synthetic data training.

3. The method according to claim 2, characterized in that The over-segmentation prediction model obtained by training based on over-segmentation synthetic data includes: Obtain historical real image data; Obtaining segmented annotation data based on the historical real image data and a preset image segmentation model; Performing generalization processing based on the over-segmented labeled data to obtain the over-segmented synthetic data; The over-segmentation prediction model is obtained based on the over-segmentation synthetic data training.

4. The method according to claim 1, characterized in that: The performing transfer learning on the over-segmentation prediction model based on part of the real data to be labeled and the segmentation labeling result to obtain the segmentation labeling model comprises: The pre-trained over-segmentation prediction model is used as a teacher model, and the teacher model is distilled based on part of the real data to be labeled to obtain the segmentation and labeling model.

5. The method according to claim 1, characterized in that The method of obtaining the over-segmentation prediction result based on part of the real data to be labeled and a pre-trained over-segmentation prediction model includes: inputting part of the real data to be labeled into the pre-trained over-segmentation prediction model, and obtaining the over-segmentation prediction result according to the model output, wherein the data amount of the part of the real data to be labeled is less than a preset threshold.

6. The method according to claim 1, characterized in that The step of merging the image segmentation blocks corresponding to the over-segmentation prediction results to obtain the segmentation annotation results corresponding to the segmentation task includes: Determining a segmentation target according to the segmentation task; The image segmentation blocks corresponding to the over-segmentation prediction results are merged according to the segmentation target to obtain a segmentation target area, and a target label is assigned to the segmentation target area to obtain the segmentation annotation result, wherein the target label includes a semantic label and / or an instance label.

7. The method according to claim 6, characterized in that The step of merging the image segmentation blocks corresponding to the over-segmentation prediction results according to the segmentation target to obtain the segmentation target region includes: determining the semantic features of the segmentation target, and merging adjacent image segmentation blocks according to the semantic features to obtain the segmentation target region.

8. A computer device comprising a processor and a memory, wherein the memory is suitable for storing a plurality of program codes, wherein: The program code is suitable for being loaded and run by the processor to execute the automatic labeling method for segmentation tasks according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a plurality of program codes, characterized in that: The program code is suitable for being loaded and run by a processor to execute the automatic labeling method for segmentation tasks according to any one of claims 1 to 7.

Images