CN-121999311-A - Labeling result verification method, electronic equipment and computer readable storage medium

CN121999311ACN 121999311 ACN121999311 ACN 121999311ACN-121999311-A

Abstract

The application discloses a marking result checking method, electronic equipment and a computer readable storage medium, wherein the method comprises the steps of obtaining a pre-marking result of a first image; the first image comprises at least one target object, the pre-labeling result is obtained by pre-labeling target behavior attributes of all target objects in the first image, a first area and a second area corresponding to target positions of all target objects are determined from the first image, the second area is an area through which the target positions of the target objects can pass under the condition that the target objects have the target behavior attributes, and the pre-labeling result of the first image is verified based on the position relation between the first area and the second area to obtain a first labeling verification result corresponding to the pre-labeling result. Through the mode, the method and the device can realize verification of the pre-labeling result.

Inventors

ZHANG XIYU
LI FENGGANG
WU LI
TANG BEIBEI
CHEN LEI

Assignees

浙江大华技术股份有限公司

Dates

Publication Date: 20260508
Application Date: 20251211

Claims (10)

1. A method for verifying a labeling result, the method comprising: the method comprises the steps of obtaining a pre-labeling result of a first image, wherein the first image comprises at least one target object, and the pre-labeling result is obtained by pre-labeling target behavior attributes of the target objects in the first image; determining a first area and a second area corresponding to a target part of each target object from the first image, wherein the second area is an area through which the target part of the target object passes when the target behavior attribute exists in the target object; And verifying the pre-labeling result of the first image based on the position relation between the first area and the second area to obtain a first labeling verification result corresponding to the pre-labeling result.
2. The method according to claim 1, wherein before verifying the pre-labeling result of the first image based on the positional relationship between the first region and the second region, and obtaining a first labeling verification result corresponding to the pre-labeling result, the method further comprises: Determining target key points in the first areas; The verifying the pre-labeling result of the first image based on the positional relationship between the first area and the second area to obtain a first labeling verification result corresponding to the pre-labeling result, including: and verifying the pre-labeling result of the first image based on the position relation between each target key point and the second region to obtain a first labeling verification result corresponding to the pre-labeling result.
3. The method of claim 2, wherein said determining target keypoints in each of said first regions comprises: Determining a preset key point percentage coordinate corresponding to the target part; and determining target key points in the first areas by utilizing the object positions of the target objects and the preset key point percentage coordinates.
4. The method of claim 2, wherein the pre-labeling results include first sub-labeling results corresponding to each target object, the first sub-labeling results corresponding to each target object being used for representing whether the target object has a target behavior attribute, and the verifying the pre-labeling results of the first image based on the positional relationship between each target key point and the second region to obtain first labeling verification results corresponding to the pre-labeling results, comprises: For each first sub-labeling result corresponding to the target object, determining that the first sub-labeling result corresponding to the target object is correct in response to whether the target key point corresponding to the target object is located in the second area and the first sub-labeling result indicates that the target object has the target behavior attribute, or whether the target key point corresponding to the target object is located outside the second area and the first sub-labeling result indicates that the target object does not have the target behavior attribute; And determining that a first sub-labeling result error corresponding to the target object exists in response to the fact that the target key point corresponding to the target object is located outside the second area and the first sub-labeling result represents that the target behavior attribute exists in the target object, or that the target key point corresponding to the target object is located in the second area and the first sub-labeling result represents that the target behavior attribute does not exist in the target object.
5. The method of claim 1, wherein the first images are at least two, the at least two first images are acquired by using an image acquisition device at the same position, and the second areas corresponding to the first images are the same: Displaying a target image on a display interface, wherein the target image is the first image or the second image, and the second image is the first image marked with the pre-marking result; And responding to the region configuration operation of the user on the display interface, and taking the region configured by the user as the second region.
6. The method of claim 5, wherein the responding to the user configuration operation in the area of the display interface, taking the area configured by the user as the second area, comprises: and responding to the region drawing operation of the user on the display interface, and taking the region drawn by the user as the second region.
7. The method according to claim 1, wherein the first images are at least two, the at least two first images are acquired by using an image acquisition device at the same position, and the second areas corresponding to the first images are the same, and after the pre-labeling result of the first images is verified based on the positional relationship between the first areas and the second areas, and the first labeling verification result corresponding to the pre-labeling result is obtained, the method further comprises: The quality verification method comprises the steps of obtaining quality verification results of a plurality of second labeling verification results, wherein the second labeling verification results are partial first labeling verification results extracted from first labeling verification results corresponding to at least two first images, and the quality verification results of the second labeling verification results are obtained by carrying out quality verification on the second labeling verification results by a user and are used for representing whether the quality of the second labeling verification results meets the standard or not; And in response to the quality check result being that the number of the second labeling check results with the quality not reaching the standard meets the number requirement, adjusting the second area and the first area of each target object, and re-executing the position relation between the first area and the second area, checking the pre-labeling result of the first image to obtain a first labeling check result corresponding to the pre-labeling result and subsequent steps until the quality check result is that the number of the second labeling check results with the quality not reaching the standard does not meet the number requirement.
8. The method of claim 1, wherein the pre-labeling results comprise first sub-labeling results for each of the target objects, the first sub-labeling results corresponding to the target objects being used to characterize whether the target object has a target behavior attribute, and wherein after verifying the pre-labeling results for the first image based on the positional relationship between each of the first region and the second region to obtain a first labeling verification result corresponding to the pre-labeling results, the method further comprises: And removing the first sub-labeling result with the first labeling check result being an error from the pre-labeling result to obtain a new pre-labeling result of the first image.
9. An electronic device comprising a memory storing program instructions and a processor for executing the program instructions to implement the labeling outcome check method of any of claims 1-8.
10. A computer readable storage medium, characterized in that the computer readable storage medium is for storing program instructions, the program instructions being executable to implement the labeling outcome check method of any of claims 1-8.

Description

Labeling result verification method, electronic equipment and computer readable storage medium Technical Field The present application relates to the field of data labeling technology, and in particular, to a labeling result verification method, an electronic device, and a computer readable storage medium. Background Image annotation has been an important task in the field of computer vision. The purpose of this task is to identify, segment and classify objects in the image to generate the final labeling result. However, due to the complexity and time-consuming nature of labeling tasks, conventional manual labeling methods have failed to meet the needs on large-scale data sets. Therefore, automated labeling and semi-automated labeling become research hotspots. The automatic labeling and the semiautomatic labeling are mainly performed by replacing manual labeling with a preset model, and the preset model has high efficiency but a certain error rate, so that the content of the preset model labeling needs to be checked. Disclosure of Invention The application mainly solves the technical problem of providing a marking result verification method, electronic equipment and a computer readable storage medium, which can realize verification of a pre-marking result. In order to solve the technical problems, the first aspect of the application provides a marking result verification method, which comprises the steps of obtaining a pre-marking result of a first image, wherein the first image comprises at least one target object, the pre-marking result is obtained by pre-marking target behavior attributes of all target objects in the first image, determining a first area and a second area corresponding to target parts of all target objects from the first image, wherein the second area is an area through which the target parts of the target objects can pass under the condition that the target behavior attributes exist in the target objects, and verifying the pre-marking result of the first image based on the position relation between the first area and the second area to obtain a first marking verification result corresponding to the pre-marking result. Before verifying the pre-labeling result of the first image based on the position relation between the first areas and the second areas to obtain a first labeling verification result corresponding to the pre-labeling result, the labeling result verification method further comprises the steps of determining target key points in the first areas, verifying the pre-labeling result of the first image based on the position relation between the first areas and the second areas to obtain a first labeling verification result corresponding to the pre-labeling result, and verifying the pre-labeling result of the first image based on the position relation between the target key points and the second areas to obtain the first labeling verification result corresponding to the pre-labeling result. The method comprises the steps of determining target key points in each first area, determining preset key point percentage coordinates corresponding to target parts, and determining the target key points in each first area by utilizing the object positions of each target object and the preset key point percentage coordinates. The pre-labeling result comprises first sub-labeling results corresponding to each target object, wherein the first sub-labeling results corresponding to the target objects are used for representing whether target behavior attributes exist in the target objects or not, the pre-labeling results of the first image are verified based on the position relation between the target key points and the second area to obtain first labeling verification results corresponding to the pre-labeling results, the first labeling verification results corresponding to the pre-labeling results comprise first sub-labeling results corresponding to the target objects, the first sub-labeling results corresponding to the target objects are responded to the first sub-labeling results, the target key points corresponding to the target objects are located in the second area, the first sub-labeling results represent the target behavior attributes, or the target key points corresponding to the target objects are located outside the second area, the first sub-labeling results represent the target objects, the first sub-labeling results corresponding to the target objects are determined to be correct, the first sub-labeling results corresponding to the target objects are determined to be incorrect, and the target key points corresponding to the target objects are located in the second area, the first sub-labeling results represent the target objects are determined to be absent. The method comprises the steps of displaying a target image on a display interface, wherein the target image is a first image or a second image, the second image is a first image marked with a pre-marking result, and the area confi