Method for generating marking result file and electronic device

Abstract

The invention provides a method for generating a marking result file and an electronic device. The method comprises the following steps of: obtaining individual marking information of data in a reference folder, and integrating the marking information of the data into a first reference marking result file; after the data is classified into a plurality of folders, obtaining the data name of each data classified into each folder, and obtaining the mark information of each data from the first reference mark result file according to the data name; and generating marking result files respectively corresponding to the plurality of folders on the basis of marking information of the data taken from the first reference marking result file.

Description

Technical Field

[0001] The present invention relates to a marking result sorting mechanism, and in particular to a method and an electronic device for generating a marking result file. Background Art

[0002] In the prior art, after the user completes the data labeling, the labeled data is generally classified into corresponding folders, and these folders are used to train, verify and test a prediction model respectively.

[0003] However, since users need to manually generate labeling result files corresponding to different folders before successfully training, verifying and testing the above prediction model, it takes a lot of manpower and time to complete the related operations. Summary of the invention

[0004] In view of this, the present invention provides a method and an electronic device for generating a marking result file, which can be used to solve the above technical problems.

[0005] An embodiment of the present invention provides a method for generating a marking result file, which is suitable for an electronic device, including: obtaining a marking information of each of a plurality of data located in a reference folder, and integrating the marking information of each data into a first reference marking result file; after the plurality of data are classified into a plurality of folders, obtaining the data name of each data classified into each folder, and obtaining the marking information of each data from the first reference marking result file based on the data name; and generating a plurality of marking result files corresponding to the plurality of folders respectively based on the marking information of each data taken from the first reference marking result file.

[0006] An embodiment of the present invention provides an electronic device, including a storage circuit and a processor. The storage circuit stores a program code. The processor is coupled to the storage circuit and accesses the program code to execute: obtaining a tag information of each of a plurality of data located in a reference folder, and integrating the tag information of each data into a first reference tag result file; after the plurality of data are classified into a plurality of folders, obtaining the data name of each data classified into each folder, and obtaining the tag information of each data from the first reference tag result file accordingly; and generating a plurality of tag result files corresponding to the plurality of folders respectively based on the tag information of each data obtained from the first reference tag result file. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] The accompanying drawings are included to provide a further understanding of the present invention and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments of the present invention and together with the description serve to explain the principles of the present invention.

[0008] Figure 1It is a schematic diagram of generating a marking result file;

[0009] Figure 2 is a schematic diagram of an electronic device according to an embodiment of the present invention;

[0010] Figure 3 is a flow chart of a method for generating a marking result file according to an embodiment of the present invention;

[0011] Figure 4 is a schematic diagram of a VIA system operating in specific software according to an embodiment of the present invention;

[0012] Figure 5 is a schematic diagram of a software interface according to an embodiment of the present invention;

[0013] Figure 6 is based on Figure 3 A flow chart of a method for generating a marking result file shown in an embodiment;

[0014] Figure 7 It is a schematic diagram of supplementary markings according to an embodiment of the present invention. DETAILED DESCRIPTION

[0015] Reference will now be made in detail to exemplary embodiments of the present invention, examples of which are illustrated in the accompanying drawings. Whenever possible, the same reference numerals are used in the drawings and the description to refer to the same or like parts.

[0016] Please refer to Figure 1 , which is a schematic diagram of generating a labeling result file. Figure 1 In the embodiment, the reference folder 10 includes, for example, a plurality of data 101 , wherein each data 101 is, for example, an image of a silicon carbide chip marked as defective by a relevant user.

[0017] Generally speaking, after a user (e.g., a data labeler) completes labeling of each data 101 using a specific labeling software, the labeling software can generate a labeling result file F1 (e.g., a COCO / JSON file) accordingly. In the first reference labeling result file F1, the relevant labeling information of each data 101 will be recorded. For example, assuming that the reference folder 10 includes data 101 with data names such as "1.jpg", "2.jpg", ..., "84.jpg", etc., then the first reference labeling result file F1 will record all the labeling information (e.g., defect locations, etc.) in each data 101 such as "1.jpg", "2.jpg", ..., "84.jpg", etc.

[0018] Afterwards, the user can classify the data 101 into folders 111 to 113 according to the needs, wherein each folder 111 to 113 can correspond to a different purpose. For example, folder 111 is a training folder, and data 101a in the data 101 classified into folder 111 can be used to train a certain prediction model (e.g., a model for defects on silicon carbide chip images). Folder 112 is a verification folder, and data 101b in the data 101 classified into folder 112 can be used to verify the above prediction model. Folder 113 is a test folder, and data 101c in the data 101 classified into folder 113 can be used to test the above prediction model.

[0019] In order to enable each folder 111-113 to achieve the corresponding purpose, the user needs to manually edit the first reference labeling result files F11-F13 (which are COCO / JSON files, respectively) corresponding to each folder 111-113. For example, for data 101a classified into folder 111, the user needs to first find the relevant labeling information corresponding to data 101a in the first reference labeling result file F1, and then manually copy the found labeling information to the first reference labeling result file F11 corresponding to folder 111.

[0020] Therefore, assuming that the data names in the reference folder 101 are "1.jpg", "4.jpg", "7.jpg", etc., and are classified into the folder 111, the user needs to manually find the marking information corresponding to the data names "1.jpg", "4.jpg", "7.jpg", etc. from the first reference marking result file F1, and manually copy them one by one to the first reference marking result file F11 corresponding to the folder 111.

[0021] Similarly, for the data 101b and 101c classified into the folders 112 and 113, the user also needs to manually copy the relevant marking information from the first reference marking result file F1 to the marking result files F12 and F13 corresponding to the folder 111 one by one.

[0022] It can be seen that the prior art requires a lot of manpower and time to generate the first reference marking result files F11 - F13 .

[0023] In view of this, an embodiment of the present invention proposes a technical solution that can be used to solve the above technical problems, and the relevant details are described as follows.

[0024] Please refer to Figure 2, which is a schematic diagram of an electronic device according to an embodiment of the present invention. In different embodiments, the electronic device 200 can be implemented as various intelligent devices and / or computer devices, or a device dedicated to marking training data and training a prediction model with the training data, but is not limited thereto.

[0025] exist Figure 2 In the embodiment, the electronic device 200 includes a storage circuit 202 and a processor 204 .

[0026] The storage circuit 202 is, for example, any type of fixed or removable random access memory (RAM), read-only memory (ROM), flash memory, hard disk or other similar device or a combination of these devices, and can be used to record multiple program codes or modules.

[0027] The processor 204 is coupled to the storage circuit 202 and can be a general-purpose processor, a special-purpose processor, a conventional processor, a digital signal processor, a plurality of microprocessors, one or more microprocessors combined with a digital signal processor core, a controller, a microcontroller, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), any other type of integrated circuit, a state machine, an Advanced RISC Machine (ARM)-based processor, and the like.

[0028] In an embodiment of the present invention, the processor 204 can access the modules and program codes recorded in the storage circuit 202 to implement the method for generating a marking result file proposed by the present invention, the details of which are described as follows.

[0029] Please refer to Figure 3 , which is a flow chart of a method for generating a marking result file according to an embodiment of the present invention. The method of this embodiment can be Figure 2 The electronic device 200 executes the following: Figure 2 Description of components shown Figure 3 In addition, for easier understanding, the following is supplemented by Figure 1 The situation is taken as an illustrative example, but possible implementations of the present invention are not limited to this.

[0030] In one embodiment, the electronic device 200 may, for example, run specific software that allows a user to label data and use the labeled data to train / validate / test the considered prediction model (hereinafter referred to as M).

[0031] In one embodiment, the specific software may be integrated with a Visual Geometry Group (VGG) image annotator (VIA) system, so that the user can perform the required marking on each data 101 through the VIA system in the specific software.

[0032] Please refer to Figure 4 , which is a schematic diagram of a VIA system operating in a specific software according to an embodiment of the present invention. In this embodiment, a user can, for example, Figure 4 The VIA system 400 shown is used to import data 410 to be tagged (e.g. Figure 1 A certain data 101 in the reference folder 10 is referenced and the required marking is made on the data 410.

[0033] For example, assuming that data 410 is a silicon carbide chip image, the user can use the VIA system 400 to mark the portion of the image corresponding to the chip defect, and the VIA system 400 will display the corresponding marking result on the image (such as marking result 411, which may include the location and range of a defect).

[0034] Furthermore, the user may also export the relevant marking information of the data 410 (such as the position and range of each defect marked on the data 410) through the VIA system 400. In one embodiment, after completing the marking of a batch of data, the user may export the marking information of the entire batch of data into a marking result file.

[0035] For example, assuming that after the user completes marking of each data 101 in the reference folder 10 through the VIA system 400, the user can export the marking information of each data 101 into a marking result file through the VIA system 400. Accordingly, the processor 104 can execute step S310 to obtain the individual marking information of the plurality of data 101 in the reference folder 10, and integrate the marking information of each data 101 into the first reference marking result file F1.

[0036] In the embodiment of the present invention, the tag information corresponding to each data 101 will be recorded in the first reference tag result file F1 together with the data name of each data 101 .

[0037] Afterwards, the user may classify the data 101 into folders 111 - 113 in the above-mentioned specific software according to requirements. For ease of explanation, it is assumed below that data 101 named "1.jpg", "4.jpg", "7.jpg", etc. in reference folder 10 are classified into folder 111 (i.e., folder 111 includes data 101a named "1.jpg", "4.jpg", "7.jpg"); data 101 named "2.jpg", "5.jpg", "8.jpg", etc. are classified into folder 112 (i.e., folder 112 includes data 101b named "2.jpg", "5.jpg", "8.jpg"); data 101 named "3.jpg", "6.jpg", "9.jpg", etc. are classified into folder 113 (i.e., folder 113 includes data 101c named "3.jpg", "6.jpg", "9.jpg", but this is only used as an example and is not used to limit possible implementation methods of the present invention.

[0038] In step S320 , after the data 101 is classified into the folders 111 - 113 , the processor 104 obtains the data name of each data 101 classified into each folder 111 - 113 , and accordingly obtains the tag information of each data 101 from the first reference tag result file F1 .

[0039] Please refer to Figure 5 , which is a schematic diagram of a software interface according to an embodiment of the present invention. Figure 5 In the example, it is assumed that the software interface 500 of the specific software includes labels 511-513 corresponding to the folders 111-113, respectively (the labels are, for example, labels named "Training", "Testing" and "Validation"), and the user selects the label 511 corresponding to the folder 111. In this case, the software interface 500 can display the data 101a classified into the folder 111.

[0040] In one embodiment, the user can delete unnecessary files in the software interface 500. In another embodiment, if some data 101a are not marked, the processor 104 can adjust the unmarked data 101a to have a thicker frame, such as Figure 5 As shown, but not limited to this.

[0041] Afterwards, the user can, for example, import the first reference labeling result file F1 by pressing a button 520 (which is, for example, a button named “Add COCO File”). In this case, the processor 104 can correspondingly obtain the data name of each data 101a-101c in each folder 111-113, and accordingly obtain the labeling information of each data 101a-101c from the first reference labeling result file F1.

[0042] Continuing with the above example, assume that the folder 111 includes data 101a named "1.jpg", "4.jpg", and "7.jpg", the folder 112 includes data 101b named "2.jpg", "5.jpg", and "8.jpg", and the folder 113 includes data 101c named "3.jpg", "6.jpg", and "9.jpg". In this case, the user can import the first reference tag result file F1 by pressing the button 520, and the processor 104 can correspondingly find the corresponding tag information in the first reference tag result file F1 according to "1.jpg", "4.jpg", and "7.jpg", find the corresponding tag information in the first reference tag result file F1 according to "2.jpg", "5.jpg", and "8.jpg", and find the corresponding tag information in the first reference tag result file F1 according to "3.jpg", "6.jpg", and "9.jpg".

[0043] For ease of understanding, the tag information corresponding to "1.jpg", "4.jpg", and "7.jpg" will be referred to as tag information LI1, LI4, and LI7, the tag information corresponding to "2.jpg", "5.jpg", and "8.jpg" will be referred to as tag information LI2, LI5, and LI8, and the tag information corresponding to "3.jpg", "6.jpg", and "9.jpg" will be referred to as tag information LI3, LI6, and LI9, but it is not limited to this.

[0044] Afterwards, in step S330 , the processor 104 generates marking result files F11 ˜ F13 corresponding to the folders 111 ˜ 113 respectively based on the marking information of each data 101 obtained from the first reference marking result file F1 .

[0045] In one embodiment, the processor 104 may generate a marking result file F11 corresponding to the folder 111 based on the marking information of each data 101a taken from the first reference marking result file F1. In this case, the marking result file F11 corresponding to the folder 111 may include the marking information LI1, LI4, and LI7. In other words, the user does not need to manually copy the marking information LI1, LI4, and LI7 from the first reference marking result file F1 to the marking result file F11, but the processor 104 may automatically find the corresponding marking information from the first reference marking result file F1 according to the data name of each data 101a, and generate the marking result file F11 corresponding to the folder 111 accordingly.

[0046] In addition, the processor 104 may also generate a marking result file F12 corresponding to the folder 112 based on the marking information of each data 101b taken from the first reference marking result file F1. In this case, the marking result file F12 corresponding to the folder 112 may include the marking information LI2, LI5, and LI8. In other words, the user does not need to manually copy the marking information LI2, LI5, and LI8 from the first reference marking result file F1 to the marking result file F12, but the processor 104 can automatically find the corresponding marking information from the first reference marking result file F1 according to the data name of each data 101b, and generate the marking result file F12 corresponding to the folder 112 accordingly.

[0047] Similarly, the processor 104 can also generate a marking result file F13 corresponding to the folder 113 based on the marking information of each data 101c taken from the first reference marking result file F1. In this case, the marking result file F13 corresponding to the folder 113 can include the marking information LI3, LI6, and LI9. In other words, the user does not need to manually copy the marking information LI3, LI6, and LI9 from the first reference marking result file F1 to the marking result file F13, but the processor 104 can automatically find the corresponding marking information from the first reference marking result file F1 according to the data name of each data 101c, and generate the marking result file F13 corresponding to the folder 113 accordingly.

[0048] It can be seen that the technical solution proposed in the embodiment of the present invention can effectively reduce the manpower and time for producing the marking result files F11-F13 corresponding to different folders 111-113.

[0049] Please refer to Figure 6 , which is based on Figure 3 The method flow chart of generating the marking result file shown in the embodiment. Figure 6In the embodiment, after executing step S330 to obtain the labeling result files F11-F13, the processor 104 may execute step S610 to train the prediction model M based on the labeling result files F11-F13, and use the trained prediction model M to predict the reference data to generate prediction results for each reference data.

[0050] In an embodiment of the present invention, the prediction model M can be implemented as various neural networks, deep learning networks and / or artificial intelligence models, and its related training parameters / training weights / training duration and other factors can be set / configured by the user in the above-mentioned specific software, but is not limited to this.

[0051] In this case, the processor 104 can train the prediction model M using the labeled result files F11 - F13 and the corresponding data 101 a - 101 c according to the user's settings.

[0052] In one embodiment, each data 101 is assumed to be a silicon carbide chip image marked with defects. In this case, when the trained prediction model M receives relevant reference data (e.g., other silicon carbide chip images that have not been identified / predicted), the prediction model M can predict the location and / or range of each defect in the reference data as the prediction result corresponding to the reference data, but is not limited thereto.

[0053] In one embodiment, the processor 104 may store the prediction results corresponding to each reference data as corresponding COCO / JSON files, and may import the prediction results corresponding to each reference data and the COCO / JSON files into the VIA system 400 for further editing by the user.

[0054] Please refer to Figure 7 , which is a schematic diagram of supplementary markings according to an embodiment of the present invention. Figure 7 In the example, the processor 104 may present the relevant prediction results in the VIA system 400 according to the COCO / JSON file corresponding to the reference data 710 .

[0055] Depend on Figure 7 It can be seen that some defects (such as defect 799 ) in the reference data 710 are not identified by the prediction model M. In this case, the user can further mark the reference data 710 through the VIA system 400 .

[0056] In other words, the prediction model M can be understood as completing a portion of the labeling of the reference data 710 , and the user can further perform additional labeling on the reference data 710 to improve the labeling behavior of the reference data 710 .

[0057] In this case, the prediction result of the reference data 710 can be understood as the (complete) labeling information of the reference data 710 modified by the user.

[0058] For other reference data, the processor 104 can also present the corresponding labeling results in the VIA system 400 according to the corresponding COCO file / JSON file for the user to modify, thereby generating labeling information corresponding to each reference data.

[0059] In step S620 , in response to determining that the prediction results of each reference data are corrected to the label information of each reference data, the processor 104 integrates the label information of each reference data with the label information of each data 101 into a second reference label result file (hereinafter referred to as F2 ).

[0060] From another perspective, after the above modification, each reference data can be regarded as new data 101 and imported into the reference folder 10, and the processor 104 can perform the behavior similar to step S310 again to generate the corresponding second reference marking result file F2. In this case, the second reference marking result file F2 includes not only the marking information of each original data 101, but also the marking information of each reference data.

[0061] For ease of explanation, it is assumed below that the data names of the reference data are "R1.jpg", "R2.jpg", and "R3.jpg", and their corresponding tag information are RI1, RI2, and RI3, but it is not limited thereto.

[0062] Afterwards, as mentioned above, the user can also classify the reference data into folders 111 to 113 according to the needs. For the convenience of explanation, it is assumed that "R1.jpg", "R2.jpg", and "R3.jpg" are classified into folders 111 to 113 respectively. In other words, folder 111 includes data 101a named "1.jpg", "4.jpg", and "7.jpg" and reference data named "R1.jpg", folder 112 includes data 101b named "2.jpg", "5.jpg", and "8.jpg" and reference data named "R2.jpg", and folder 113 includes data 101c named "3.jpg", "6.jpg", and "9.jpg" and reference data named "R3.jpg".

[0063] Next, in step S630, after the reference data is classified into folders 111 to 113, the processor 104 obtains the data name of each data 101 classified into each folder 111 to 113 and the data name of each reference data, and accordingly obtains the marking information of each data 101 and the marking information of each reference data from the second reference marking result file F2.

[0064] In one embodiment, the user can import the second reference tag result file F2 by pressing button 520, and the processor 104 can correspondingly find the corresponding tag information in the second reference tag result file F2 based on "1.jpg", "4.jpg", "7.jpg" and "R1.jpg", find the corresponding tag information in the second reference tag result file F2 based on "2.jpg", "5.jpg", "8.jpg" and "R2.jpg", and find the corresponding tag information in the second reference tag result file F2 based on "3.jpg", "6.jpg", "9.jpg" and "R3.jpg".

[0065] Afterwards, in step S640 , the processor 104 generates a plurality of new marking result files corresponding to the folders 111 - 113 respectively based on the marking information of each data 101 obtained from the second reference marking result file F2 and the marking information of each reference data.

[0066] In one embodiment, the processor 104 may generate a new marking result file (hereinafter referred to as F11') corresponding to the folder 111 based on the marking information of each data 101a taken from the second reference marking result file F2 and RI1. In this case, the new marking result file F11' corresponding to the folder 111 may include marking information LI1, LI4, LI7 and RI1.

[0067] In addition, the processor 104 may generate a new marking result file (hereinafter referred to as F12') corresponding to the folder 112 based on the marking information of each data 101b taken from the second reference marking result file F2 and RI2. In this case, the new marking result file F12' corresponding to the folder 112 may include marking information LI2, LI5, LI8 and RI2.

[0068] Similarly, the processor 104 may generate a new marking result file (hereinafter referred to as F13') corresponding to the folder 113 based on the marking information of each data 101c taken from the second reference marking result file F2 and RI3. In this case, the new marking result file F13' corresponding to the folder 113 may include marking information LI3, LI6, LI9 and RI9.

[0069] Afterwards, in step S650 , the processor 104 may train the prediction model M based on the new labeling result files F11 ′ to F13 ′.

[0070] As can be seen from the above, after the user corrects the prediction results of each reference data into corresponding marking information, the processor 104 can update the marking result files F11 ′- F13 ′ corresponding to each folder 111 - 113 and train the prediction model M accordingly.

[0071] From another perspective, the prediction model M can be used to assist the user in making a part of the prediction / labeling for the unlabeled reference data, thereby increasing the efficiency of the user's labeling. Moreover, after the user corrects the prediction results of the reference data into a more perfect labeling information, these labeling information can be used together with the existing labeling information to further train the prediction model M. Thereby, the prediction accuracy of the prediction model M can be improved.

[0072] To make the above effects easier to understand, Tables 1 and 2 are supplemented below for illustration. Table 1 shows the results of labeling in the existing manner, while Table 2 shows the results of labeling in the manner of this case. Among them, the overall accuracy can be characterized as, for example, "(actual number of defects - number of missed defects - number of mis-captured defects) / actual number of defects", but it is not limited thereto. In addition, in Table 2, Version 1 is manual labeling (control group); Version 2 is to train the prediction model M using the data of Version 1, and after automatically labeling with the prediction model M, then manually label the remaining data; Version 3 is to train the prediction model M using the data of Version 2, and after automatically labeling with the prediction model M, then manually label the remaining data.

[0073]

[0074] Table 1 (Results of labeling in the existing manner)

[0075]

[0076] Table 2 (Results of labeling in the manner of this case)

[0077] In Tables 1 and 2, it is assumed that the relevant personnel need an average of 7 minutes to complete the labeling of one image. In Version 1 of Table 1, the relevant personnel need about 1540 minutes to complete the labeling of 220 images. In Version 2 of Table 1, a total of 270 images need to be labeled, but after removing the 220 images already labeled in Version 1, there are 50 remaining images to be labeled, and the relevant personnel need about 350 minutes to complete the labeling of 50 images. In Version 3 of Table 1, a total of 430 images need to be labeled, but after removing the 270 images already labeled in Version 2, there are 160 remaining images to be labeled, and the relevant personnel need about 1120 minutes to complete the labeling of 160 images.

[0078] In contrast, in the second version of Table 2, since the electronic device 200 can accurately and automatically label 74% of the 50 images through the trained prediction model M, the relevant personnel only need to label 26% (i.e., 100%-74%) of the 50 images, and thus only need 91 minutes (which is much lower than the 350 minutes required for the second version of Table 1) to complete the labeling. In other words, the relevant personnel only need to label the 26% of images that the prediction model M failed to correctly predict in the first version.

[0079] Furthermore, in version 3 of Table 2, since the electronic device 200 can accurately and automatically label 80% of the 160 images through the trained prediction model M, the relevant personnel only need to label 20% (i.e., 100%-80%) of the 160 images, and thus only need 224 minutes (which is much lower than the 1120 minutes required for version 3 of Table 1) to complete the labeling. In other words, the relevant personnel only need to label the 20% of images that the prediction model M failed to correctly predict in version 2.

[0080] It can be seen that the method proposed in the embodiment of the present invention can effectively reduce the time used for marking images.

[0081] In summary, the technical solution proposed in the embodiment of the present invention can automatically find the corresponding marking information from the first reference marking result file according to the data name of each data located in different folders, and generate marking result files corresponding to different folders accordingly. Thereby, the manpower and time for producing marking result files corresponding to different folders can be effectively reduced.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a marking result file, suitable for an electronic device, characterized in that: include: Obtaining individual marking information of a plurality of data located in a reference folder, and integrating the marking information of each of the data into a first reference marking result file; After the plurality of data are classified into a plurality of folders, obtaining a data name of each of the data classified into each of the folders, and obtaining the tag information of each of the data from the first reference tag result file accordingly; as well as A plurality of marking result files respectively corresponding to the plurality of folders are generated based on the marking information of each of the data taken from the first reference marking result file.

2. The method according to claim 1, wherein the plurality of folders include a first folder, and the step of obtaining the data name of each of the data classified into each of the folders and obtaining the tag information of each of the data from the first reference tag result file accordingly comprises: The first data name of each of at least one first data classified into the first folder is obtained, and the tag information of each first data is obtained from the first reference tag result file accordingly.

3. The method according to claim 2, wherein the plurality of marking result files include a first marking result file corresponding to the first folder, and the step of generating the plurality of marking result files corresponding to the plurality of folders respectively based on the marking information of each of the data taken from the first reference marking result file comprises: The first marking result file corresponding to the first folder is generated based on the marking information of each of the first data taken from the first reference marking result file.

4. The method according to claim 1, wherein after the step of generating the plurality of marking result files respectively corresponding to the plurality of folders, the method further comprises: Training a prediction model based on the multiple labeled result files, and using the trained prediction model to predict at least one reference data to generate a prediction result for each reference data; In response to determining that the prediction result of each of the reference data is corrected to the tag information of each of the reference data, the tag information of each of the reference data and the tag information of each of the data are integrated into a second reference tag result file.

5. The method according to claim 4, further comprising: After the plurality of reference data are classified into the plurality of folders, the data name of each of the data classified into each of the folders and the data name of each of the reference data are obtained, and accordingly the tag information of each of the data and the tag information of each of the reference data are obtained from the second reference tag result file; generating a plurality of new marking result files respectively corresponding to the plurality of folders based on the marking information of each of the data taken from the second reference marking result file and the marking information of each of the reference data; as well as The prediction model is trained based on the multiple new labeled result files. The method according to claim 1 , wherein the plurality of folders include a training folder, a testing folder, and a validation folder. 7 . The method according to claim 1 , wherein each of the data comprises a silicon carbide chip image, and the marking information of each of the data represents at least one defect on the corresponding silicon carbide chip image. 8 . The method according to claim 1 , wherein the first reference labeling result file and each of the labeling result files are respectively COCO files or JSON files.

9. An electronic device, characterized in that: include: a storage circuit storing program code; as well as A processor coupled to the storage circuit and accessing the program code to execute: Obtaining individual marking information of a plurality of data located in a reference folder, and integrating the marking information of each of the data into a first reference marking result file; After the plurality of data are classified into a plurality of folders, obtaining a data name of each of the data classified into each of the folders, and obtaining the tag information of each of the data from the first reference tag result file accordingly; as well as A plurality of marking result files respectively corresponding to the plurality of folders are generated based on the marking information of each of the data taken from the first reference marking result file.

10. The electronic device of claim 9, wherein the plurality of folders includes a first folder, and the processor is configured to: The first data name of each of at least one first data classified into the first folder is obtained, and the tag information of each first data is obtained from the first reference tag result file accordingly.

Images