Automatic data labeling method and system

Through automated data labeling methods and systems, the problem of inefficient data collection and model update in image data labeling is solved, automated data cleaning and model update are realized, and data labeling efficiency is improved.

CN118942089APending Publication Date: 2024-11-12HANGZHOU JIEFENG SOFTWARE CO LTD
View PDF 12 Cites 0 Cited by

Patent Information

Application Number
CN202410996043.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The prior art relies on manual operations in the process of large-scale image data annotation, especially in the data collection, cleaning and model update stages, resulting in inefficiency and no automation is achieved.

Method used

Through automatic labeling methods and systems, collect data, perform deduplication and abnormal image culling, image labeling and model training, and realize automated data cleaning and model update. The specific steps include obtaining images that are suitable for the needs of the scenario, performing multimodal model similarity calculation, deduplication, grayscale standard deviation calculation, labeling and model training to ensure data quality and model accuracy.

Benefits of technology

The data labeling process is automated, the efficiency of data cleaning and model update is improved, manual intervention is reduced, and the overall efficiency of data labeling is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118942089A_ABST
    Figure CN118942089A_ABST
Patent Text Reader

Abstract

The invention provides an automatic data labeling method and system, and the method comprises the steps: obtaining collection data, and obtaining various images suitable for scene demands according to the collection data; performing duplicate removal on various images suitable for scene requirements to obtain duplicate-removed images; removing an abnormal image from the de-duplicated image to obtain a cleaned image; labeling the cleaned image on an image labeling system to obtain a labeled data set; training a pre-marking model by using the marked data set to obtain the pre-marking model meeting the requirement; integrating the pre-marking model meeting the requirement on an image marking system; automatic data cleaning and model updating can be realized, and the efficiency of data annotation is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an automatic data annotation method and system. Background Art

[0002] At present, large-scale image data annotation is the basis for visual algorithms to achieve good performance. Image data annotation has three stages: collecting required data, data cleaning, and image target labeling.

[0003] Currently, data annotation is done manually or semi-automatically. Semi-automation is mainly done in the image target labeling stage. In this stage, a model is used to obtain the label of the required target, and then manual review is performed. Among them, manual review includes adding, deleting or modifying labels.

[0004] However, there is no automation in terms of collecting the required data and cleaning the data, and manual updates are also required during the model update phase. This leads to low efficiency when large-scale data annotation is required. Summary of the invention

[0005] In view of this, an object of the present invention is to provide a method and system for automatic data labeling, which can realize automatic data cleaning and model updating, and improve the efficiency of data labeling.

[0006] In a first aspect, an embodiment of the present invention provides a method for automatically labeling data, the method comprising:

[0007] Acquire collected data, and acquire images applicable to various scene requirements according to the collected data;

[0008] Deduplication of images required by various applicable scenarios is performed to obtain deduplicated images;

[0009] Eliminate abnormal images from the deduplicated images to obtain a cleaned image;

[0010] Annotating the cleaned image on an image annotation system to obtain an annotated data set;

[0011] Using the labeled data set to train the pre-labeling model, to obtain a pre-labeling model that meets the requirements;

[0012] Integrate the pre-labeled model that meets the requirements into the image annotation system.

[0013] Furthermore, images suitable for various scenarios are obtained based on the collected data, including:

[0014] Acquire the various applicable scenarios, each of which includes at least one text description;

[0015] Inputting the text description and the image in the collected data into a multimodal model to obtain similarity;

[0016] If the similarity is less than or equal to a first preset threshold, removing the image from the image library;

[0017] If the similarity is greater than the first preset threshold, the image is retained and used as the image applicable to the scene requirement.

[0018] Furthermore, deduplication is performed on the images required for the various applicable scenarios to obtain deduplicated images, including: repeatedly performing the following processing until each of the images is traversed:

[0019] Selecting a first image and a second image from the images required by the various applicable scenarios;

[0020] Divide the first image into M*N blocks;

[0021] Divide the second image into the M*N blocks;

[0022] Compare the M*N blocks of the first image with the blocks at corresponding positions in the M*N blocks of the second image to obtain M*N similarities;

[0023] Averaging the M*N similarities to obtain an average value;

[0024] When the average value is greater than a second preset threshold, the first image or the second image is retained.

[0025] Further, removing abnormal images from the deduplicated images to obtain cleaned images includes:

[0026] Calculate the grayscale mean and standard deviation of the current image;

[0027] When the standard deviation is less than or equal to a third preset threshold, taking the current image as the abnormal image;

[0028] When the average value is greater than a first brightness threshold or the average value is less than a second brightness threshold, taking the current image as the abnormal image;

[0029] After removing the abnormal image, the cleaned image is obtained;

[0030] Wherein, the first brightness threshold is greater than the second brightness threshold.

[0031] Furthermore, the cleaned image is annotated on an image annotation system to obtain an annotated data set, including:

[0032] marking the cleaned image on the image annotation system and outputting a plurality of export data;

[0033] Decompressing the plurality of export data respectively to obtain a plurality of decompressed export data, and saving the plurality of decompressed export data in a set location;

[0034] Based on the plurality of decompressed exported data, different label information corresponding to each identical image is automatically merged to obtain the annotated data set.

[0035] Furthermore, the pre-labeling model is trained using the labeled data set to obtain a pre-labeling model that meets the requirements, including:

[0036] Set N1 training rounds, where N1 is greater than or equal to 2;

[0037] The pre-labeled model after the second round of training is tested on the validation set to obtain the second round of test results;

[0038] Comparing the second round of test results with the first round of test results;

[0039] When the accuracy of the second round of test results is greater than the accuracy of the first round of test results, retaining the pre-labeled model after the second round of training, and removing the pre-labeled model after the first round of training;

[0040] When the accuracy of the second round of test results is less than the accuracy of the first round of test results, retaining the pre-labeled model after the first round of training, and removing the pre-labeled model after the second round of training;

[0041] When the N1th round of training is reached, the pre-labeled model after the N1th round of training is tested using the validation set to obtain the N1th round of test results;

[0042] Compare the N1th round test result with the N1-1th round test result;

[0043] When the accuracy of the N1th round test result is greater than the accuracy of the N1-1th round test result, retain the pre-labeled model after the N1th round training, remove the pre-labeled model after the N1-1th round training, and use the pre-labeled model after the N1th round training as the pre-labeled model that meets the requirements;

[0044] When the accuracy of the N1th round of test results is less than the accuracy of the N1-1th round of test results, the pre-labeled model after the N1-1th round of training is retained, the pre-labeled model after the N1th round of training is removed, and the pre-labeled model after the N1-1th round of training is used as the pre-labeled model that meets the requirements.

[0045] Furthermore, the method further comprises:

[0046] Get test data;

[0047] After the test data is imported into the image annotation system, the pre-labeling model that meets the requirements is started;

[0048] The image annotation system inputs the test data into the pre-labeling model that meets the requirements to obtain the detection result.

[0049] In a second aspect, an embodiment of the present invention provides an automatic data annotation system, the system comprising:

[0050] An acquisition module is used to acquire collected data and obtain images suitable for various scene requirements according to the collected data;

[0051] A deduplication module, used to deduplicate images required by various applicable scenarios to obtain deduplicated images;

[0052] A removal module is used to remove abnormal images from the deduplicated images to obtain a cleaned image;

[0053] A labeling module, used for labeling the cleaned image on an image labeling system to obtain a labeled data set;

[0054] A training module, used to train the pre-labeling model using the labeled data set to obtain a pre-labeling model that meets the requirements;

[0055] An integration module is used to integrate the pre-labeling model that meets the requirements into the image annotation system.

[0056] In a third aspect, an embodiment of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor implements the method described above when executing the computer program.

[0057] The embodiment of the present invention provides an automatic data labeling method and system, including: acquiring collected data, and acquiring images required by various applicable scenarios based on the collected data; deduplicating images required by various applicable scenarios to obtain deduplicated images; removing abnormal images from the deduplicated images to obtain cleaned images; labeling the cleaned images on an image labeling system to obtain a labeled data set; training a pre-labeling model using the labeled data set to obtain a pre-labeling model that meets the requirements; integrating the pre-labeling model that meets the requirements into the image labeling system; automatic data cleaning and model updating can be achieved, thereby improving the efficiency of data labeling.

[0058] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0059] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0061] Figure 1 A flow chart of the automatic data annotation method provided in the first embodiment of the present invention;

[0062] Figure 2 A schematic diagram of an image deduplication process provided in the first embodiment of the present invention;

[0063] Figure 3 A schematic diagram of a process for merging multiple exported data provided in the first embodiment of the present invention;

[0064] Figure 4 A schematic diagram of the model automatic training deployment process provided in the first embodiment of the present invention;

[0065] Figure 5 A schematic diagram of an automatic data annotation system provided in Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] To facilitate understanding of this embodiment, the embodiment of the present invention is described in detail below.

[0068] Embodiment 1:

[0069] Figure 1 This is a flow chart of the automatic data annotation method provided in Example 1 of the present invention.

[0070] Reference Figure 1 , the method comprises the following steps:

[0071] Step S101, acquiring collected data, and acquiring images suitable for various scene requirements according to the collected data;

[0072] Step S102, deduplicating images required by various applicable scenarios to obtain deduplicated images;

[0073] Step S103, removing abnormal images from the deduplicated images to obtain a cleaned image;

[0074] Step S104, annotating the cleaned image on an image annotation system to obtain an annotated data set;

[0075] Step S105, training the pre-labeling model using the labeled data set to obtain a pre-labeling model that meets the requirements;

[0076] Step S106: Integrate the pre-labeled model that meets the requirements into the image annotation system.

[0077] In this application, the automatic data labeling method includes data collection, data cleaning, and update deployment of the target labeling model. In the process of data collection, it is necessary to collect data from multiple channels to ensure that the collected data is the data required for training the pre-labeling model; then, images of various applicable scene requirements are obtained based on the collected data.

[0078] The images required by various applicable scenarios are cleaned, and the cleaning process includes deduplication and removal of abnormal images. That is, the images required by various applicable scenarios are deduplicated to obtain deduplicated images; abnormal images are removed from the deduplicated images to obtain cleaned images.

[0079] In order to improve the labeling speed, it is necessary to continuously improve the capabilities of the pre-labeling model, that is, the pre-labeling model needs to be continuously updated to improve the accuracy of the model labeling. The pre-labeling model is trained with the labeled data set to obtain a pre-labeling model that meets the requirements; the pre-labeling model that meets the requirements is integrated into the image labeling system.

[0080] Furthermore, step S101 includes the following steps:

[0081] Step S201, obtaining various applicable scenarios, each of which includes at least one text description;

[0082] Here, various applicable scenes include outdoor, indoor, store, weather and other scenes. For example, people indoors and cats outdoors. Each applicable scene includes at least one text description. For example, it is raining outdoors, there are people in a shopping mall, and a cat on a sofa.

[0083] Step S202, inputting the text description and the image in the collected data into a multimodal model to obtain similarity;

[0084] Specifically, the multimodal model is used to calculate the similarity between the image and the text, referring to formula (1):

[0085] score = MM (image, text)

[0086] Among them, MM is a multimodal model, image is an image, text is a text, and score is the similarity.

[0087] Step S203, if the similarity is less than or equal to the first preset threshold, remove the image from the image library;

[0088] Step S204: if the similarity is greater than the first preset threshold, the image is retained and used as an image suitable for scene requirements.

[0089] Specifically, if the similarity score>S t , where S t If it is the first preset threshold, it means that the image is a required image and is retained; otherwise, it is an unnecessary image and is removed from the image library.

[0090] Further, step S102 includes: repeatedly performing the following process until each image is traversed:

[0091] Step S301, selecting a first image and a second image from various images applicable to scene requirements;

[0092] Step S302, dividing the first image into M*N blocks;

[0093] Step S303, dividing the second image into the M*N blocks;

[0094] Step S304, comparing the M*N blocks of the first image with the blocks at corresponding positions in the M*N blocks of the second image to obtain M*N similarities;

[0095] Step S305, averaging the M*N similarities to obtain an average value;

[0096] Step S306: when the average value is greater than the second preset threshold, retain the first image or the second image.

[0097] Specifically, refer to Figure 2, compare the first image and the second image; divide the first image into M*N blocks, and divide the second image into M*N blocks; compare the blocks at corresponding positions of the first image and the second image one by one, so as to obtain M*N similarities; average the M*N similarities and calculate the average value; compare the average value with a second preset threshold; when the average value is greater than the second preset threshold, it means that the first image and the second image are similar, and at this time, one of the images can be retained.

[0098] Further, step S103 includes the following steps:

[0099] Step S401, calculating the grayscale average and standard deviation of the current image;

[0100] Step S402, when the standard deviation is less than or equal to a third preset threshold, the current image is regarded as an abnormal image;

[0101] Step S403, when the average value is greater than the first brightness threshold or the average value is less than the second brightness threshold, the current image is regarded as an abnormal image;

[0102] Step S404, after removing the abnormal image, a cleaned image is obtained;

[0103] The first brightness threshold is greater than the second brightness threshold.

[0104] Specifically, the elimination of abnormal data mainly involves eliminating images that are too dark, too bright, and too blurred. This application uses image processing technology to eliminate abnormal data. Calculate the grayscale average and standard deviation of the current image; when the standard deviation is less than or equal to the third preset threshold, it means that the current image is too blurred; when the average value is greater than the first brightness threshold, it means that the current image is too bright; when the average value is less than the second brightness threshold, it means that the current image is too dark. Then, based on these judgment results, eliminate abnormal data.

[0105] Further, step S104 includes the following steps:

[0106] Step S501, marking the cleaned image on the image annotation system, and outputting a plurality of export data;

[0107] Step S502, decompressing the plurality of export data respectively to obtain a plurality of decompressed export data, and saving the plurality of decompressed export data in a set location;

[0108] Step S503: Based on the multiple decompressed exported data, different label information corresponding to each identical image is automatically merged to obtain a labeled data set.

[0109] Specifically, in order to improve the labeling speed, it is necessary to continuously improve the capabilities of the pre-labeling model, that is, the pre-labeling model needs to be continuously updated to improve the accuracy of the model labeling. The implementation process of the image annotation system for model update and deployment is: mark the cleaned images on the image annotation system and output multiple export data. Use the interface provided by the image annotation system to automatically export the corresponding data to the specified location; after the export is completed, the program automatically decompresses the exported data and saves it to the specified location. After the decompression is completed, multiple decompressed export data are automatically merged, that is, the different label information corresponding to each identical image is automatically merged to obtain the labeled data set. For details, refer to Figure 3 .

[0110] Finally, the data set is converted into the format required for training and the data format file required for training is generated.

[0111] Further, step S105 includes the following steps:

[0112] Step S601, setting N1 rounds of training, where N1 is greater than or equal to 2;

[0113] Step S602, testing the pre-labeled model after the second round of training through the validation set to obtain the second round of test results;

[0114] Step S603, comparing the second round test results with the first round test results;

[0115] Step S604, when the accuracy of the second round of test results is greater than the accuracy of the first round of test results, retain the pre-labeled model after the second round of training, and remove the pre-labeled model after the first round of training;

[0116] Step S605, when the accuracy of the second round of test results is less than the accuracy of the first round of test results, retain the pre-labeled model after the first round of training, and remove the pre-labeled model after the second round of training;

[0117] Step S606, when the N1th round of training is reached, the pre-labeled model after the N1th round of training is tested on the validation set to obtain the N1th round of test results;

[0118] Step S607, comparing the N1th round test result with the N1-1th round test result;

[0119] Step S608, when the accuracy of the N1th round test result is greater than the accuracy of the N1-1th round test result, retain the pre-labeled model trained in the N1th round, remove the pre-labeled model trained in the N1-1th round, and use the pre-labeled model trained in the N1th round as the pre-labeled model that meets the requirements;

[0120] Step S609, when the accuracy of the test results of the N1th round is less than the accuracy of the test results of the N1-1th round, retain the pre-labeled model after the N1-1th round of training, remove the pre-labeled model after the N1th round of training, and use the pre-labeled model after the N1-1th round of training as the pre-labeled model that meets the requirements.

[0121] Specifically, the pre-labeled model that meets the needs is selected in each training. After each training, the test results will be immediately verified on the validation set and compared with the test results of the previous round of models. The model with the highest accuracy (i.e. the best effect) will be retained.

[0122] After the model is trained, the pre-labeled model that meets the requirements is uploaded to the specified location; then the pre-labeled model that meets the requirements is obtained from the specified location and integrated into the image annotation system. The image annotation system can be a CVAT system. Figure 4 . Figure 4 In , configuration information can be set by the user, and data processing, training, and model deployment can be achieved through automation.

[0123] Furthermore, the method further comprises the following steps:

[0124] Step S801, obtaining test data;

[0125] Step S802, after the test data is imported into the image annotation system, a pre-labeling model that meets the requirements is started;

[0126] In step S803, the image annotation system inputs the test data into a pre-labeling model that meets the requirements to obtain a detection result.

[0127] Specifically, after the pre-labeling model that meets the requirements is integrated into the image annotation system, the test data is imported into the image annotation system, the image annotation system starts the pre-labeling model that meets the requirements, the image annotation system inputs the test data into the pre-labeling model that meets the requirements, and the pre-labeling model that meets the requirements returns the detection results.

[0128] This application quickly obtains required data and completes data cleaning from a large amount of data; and uses program algorithms to automate the labeling process to the greatest extent possible to improve labeling efficiency.

[0129] Embodiment 2:

[0130] Figure 5 A schematic diagram of an automatic data annotation system provided in Embodiment 2 of the present invention.

[0131] Reference Figure 5 , the system comprises:

[0132] An acquisition module is used to acquire collected data and obtain images suitable for various scenarios based on the collected data;

[0133] The deduplication module is used to deduplicate images required by various scenarios and obtain deduplicated images;

[0134] A removal module is used to remove abnormal images from the deduplicated images to obtain a cleaned image;

[0135] An annotation module is used to annotate the cleaned images on the image annotation system to obtain an annotated data set;

[0136] A training module is used to train the pre-labeling model using the labeled data set to obtain a pre-labeling model that meets the requirements;

[0137] The integration module is used to integrate the pre-labeled model that meets the requirements into the image annotation system.

[0138] The embodiment of the present invention provides an automatic data labeling method and system, including: acquiring collected data, and acquiring images required by various applicable scenarios based on the collected data; deduplicating images required by various applicable scenarios to obtain deduplicated images; removing abnormal images from the deduplicated images to obtain cleaned images; labeling the cleaned images on an image labeling system to obtain a labeled data set; training a pre-labeling model using the labeled data set to obtain a pre-labeling model that meets the requirements; integrating the pre-labeling model that meets the requirements into the image labeling system; automatic data cleaning and model updating can be achieved, thereby improving the efficiency of data labeling.

[0139] An embodiment of the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the automatic data labeling method provided in the above embodiment are implemented.

[0140] An embodiment of the present invention further provides a computer-readable medium having a non-volatile program code executable by a processor, wherein a computer program is stored on the computer-readable medium, and when the computer program is executed by the processor, the steps of the automatic data labeling method of the above embodiment are executed.

[0141] The computer program product provided in the embodiment of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method described in the previous method embodiment. The specific implementation can be found in the method embodiment and will not be repeated here.

[0142] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system and device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0143] In addition, in the description of the embodiments of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0144] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0145] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0146] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for automatic data annotation, characterized in that: The method comprises: Acquire collected data, and acquire images applicable to various scene requirements according to the collected data; Deduplication of images required by various applicable scenarios is performed to obtain deduplicated images; Eliminate abnormal images from the deduplicated images to obtain a cleaned image; Annotating the cleaned image on an image annotation system to obtain an annotated data set; Using the labeled data set to train the pre-labeling model to obtain a pre-labeling model that meets the requirements; Integrate the pre-labeled model that meets the requirements into the image annotation system.

2. The automatic data annotation method according to claim 1, characterized in that: According to the collected data, images suitable for various scene requirements are obtained, including: Acquire the various applicable scenarios, each of which includes at least one text description; Inputting the text description and the image in the collected data into a multimodal model to obtain similarity; If the similarity is less than or equal to a first preset threshold, removing the image from the image library; If the similarity is greater than the first preset threshold, the image is retained and used as the image applicable to the scene requirement.

3. The automatic data annotation method according to claim 1, characterized in that: Deduplication of images required by various applicable scenarios is performed to obtain deduplicated images, including: repeatedly performing the following processing until each of the images is traversed: Selecting a first image and a second image from the images required by the various applicable scenarios; Divide the first image into M*N blocks; Divide the second image into the M*N blocks; Compare the M*N blocks of the first image with the blocks at corresponding positions in the M*N blocks of the second image to obtain M*N similarities; Averaging the M*N similarities to obtain an average value; When the average value is greater than a second preset threshold, the first image or the second image is retained.

4. The automatic data annotation method according to claim 1, characterized in that: Eliminating abnormal images from the deduplicated images to obtain cleaned images includes: Calculate the grayscale mean and standard deviation of the current image; When the standard deviation is less than or equal to a third preset threshold, taking the current image as the abnormal image; When the average value is greater than a first brightness threshold or the average value is less than a second brightness threshold, taking the current image as the abnormal image; After removing the abnormal image, the cleaned image is obtained; Wherein, the first brightness threshold is greater than the second brightness threshold.

5. The automatic data annotation method according to claim 1, characterized in that: The cleaned image is annotated on an image annotation system to obtain an annotated data set, including: marking the cleaned image on the image annotation system and outputting a plurality of export data; Decompressing the plurality of export data respectively to obtain a plurality of decompressed export data, and saving the plurality of decompressed export data in a set location; Based on the plurality of decompressed exported data, different label information corresponding to each identical image is automatically merged to obtain the annotated data set.

6. The automatic data annotation method according to claim 1, characterized in that: The pre-labeling model is trained using the labeled data set to obtain a pre-labeling model that meets the requirements, including: Set N1 training rounds, where N1 is greater than or equal to 2; The pre-labeled model after the second round of training is tested on the validation set to obtain the second round of test results; Comparing the second round of test results with the first round of test results; When the accuracy of the second round of test results is greater than the accuracy of the first round of test results, retaining the pre-labeled model after the second round of training, and removing the pre-labeled model after the first round of training; When the accuracy of the second round of test results is less than the accuracy of the first round of test results, retaining the pre-labeled model after the first round of training, and removing the pre-labeled model after the second round of training; When the N1th round of training is reached, the pre-labeled model after the N1th round of training is tested using the validation set to obtain the N1th round of test results; Compare the N1th round test result with the N1-1th round test result; When the accuracy of the N1th round test result is greater than the accuracy of the N1-1th round test result, retain the pre-labeled model after the N1th round training, remove the pre-labeled model after the N1-1th round training, and use the pre-labeled model after the N1th round training as the pre-labeled model that meets the requirements; When the accuracy of the N1th round of test results is less than the accuracy of the N1-1th round of test results, the pre-labeled model after the N1-1th round of training is retained, the pre-labeled model after the N1th round of training is removed, and the pre-labeled model after the N1-1th round of training is used as the pre-labeled model that meets the requirements.

7. The automatic data annotation method according to claim 1, characterized in that: The method further comprises: Get test data; After the test data is imported into the image annotation system, the pre-labeling model that meets the requirements is started; The image annotation system inputs the test data into the pre-labeling model that meets the requirements to obtain the detection result.

8. A data automatic labeling system, characterized in that: The system comprises: An acquisition module is used to acquire collected data and obtain images suitable for various scene requirements according to the collected data; A deduplication module, used to deduplicate images required by various applicable scenarios to obtain deduplicated images; A removal module is used to remove abnormal images from the deduplicated images to obtain a cleaned image; A labeling module, used for labeling the cleaned image on an image labeling system to obtain a labeled data set; A training module, used to train the pre-labeling model using the labeled data set to obtain a pre-labeling model that meets the requirements; An integration module is used to integrate the pre-labeling model that meets the requirements into the image annotation system.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer readable medium having a non-volatile program code executable by a processor, characterized in that: The program code enables the processor to execute the method according to any one of claims 1 to 7.

Citation Information

Patent Citations

  • Synthetic aperture radar target identification method for small sample residual error learning

    CN109492556A

  • A capsule endoscope image classification system and method based on transfer learning

    CN109934276A

  • Drivable area labeling method and device and electronic equipment

    CN112348122A

  • Data annotation model training method and device thereof, electronic equipment and storage medium

    CN114078578A

  • Image data annotation method and device, electronic equipment and storage medium

    CN115908977A