Data annotation method and system based on WEB end target identification

Abstract

The invention provides a data annotation method and system based on WEB end target identification, and belongs to the technical field of data annotation processing. The data standard method comprises the steps that after an image to be labeled is configured to a WEB end, the WEB end starts a pre-labeling function, a target detection model deployed in the WEB end is directly called for pre-labeling, and a pre-labeling result is obtained; the pre-labeling result is visually displayed in a to-be-labeled image of a WEB end, the use frequency of all labels in the pre-labeling result is counted and arranged in a descending order to be used for correcting the pre-labeling result, and the labels with the same frequency are arranged in an ascending order according to initial letters of names; according to the invention, the data transmission quantity and the performance requirement on the server are greatly reduced, the time consumption is shortened, the overtime failure problem is effectively avoided, and the reliability is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data annotation processing, and particularly relates to a data annotation method and system for target recognition based on the WEB side. Background Art

[0002] The statements in this part merely provide the background art related to the present invention and do not necessarily constitute the prior art.

[0003] Data annotation is a process of processing data such as pictures, voices, texts, videos, etc. through classification, framing, annotation, commenting, etc., and marking the characteristics of objects to be used as the basic materials for machine learning; for image data, the original data annotation method mostly adopts offline manual annotation, which has low work efficiency and long cycle. In view of the problems existing in manual annotation, an intelligent annotation method has been proposed in recent years. The server calls an algorithm model to implement pre-annotation, and then the user performs secondary annotation based on the pre-annotation results.

[0004] Patent No. CN115512177A discloses an artificial intelligence annotation method and annotation system based on the WEB. The pre-annotation function is started on the WEB side, and a pre-annotation request is sent to the artificial intelligence server. The artificial intelligence server receives the pre-annotation request sent by the WEB side, calls a deep learning algorithm model to pre-annotate the target object of each picture to be annotated, and returns the pre-annotation results corresponding to the picture set to be annotated to the WEB side for the user to perform secondary annotation based on the pre-annotation results.

[0005] The inventor found that most of the existing intelligent annotations adopt the method of the server calling the algorithm model. The data flow in the process of obtaining the pre-annotation results has experienced the process of "WEB side → artificial intelligence server → algorithm model → artificial intelligence server → WEB side". Affected by network performance and server resources, the response is slow, and it is easy to have problems such as request timeout and failure, and the performance is extremely unstable; in addition, during secondary annotation, only the addition of missed objects, the deletion of mislabeled objects, and the adjustment of the boundary of the annotation box are supported, and directly modifying the labels and shapes of the pre-annotation results is not supported. When modifying the labels and shapes, the method of first deleting the original mislabeled object and then adding the missed object needs to be adopted, which is a cumbersome process, and the label order is fixed, which is time-consuming to search when the number of labels is large; the workload of secondary annotation can reflect the algorithm performance. The existing solutions lack real-time statistics of the workload during secondary annotation, and manual statistics will waste a lot of manpower, material resources and financial resources and are extremely prone to errors. Summary of the Invention

[0006] To solve the deficiencies of the prior art, the present invention provides a data annotation method and system based on WEB - end target recognition. By using the method of calling a pre - annotation algorithm model on the WEB - end, the target detection model is deployed to the local WEB - end. The WEB - end directly calls the target detection model to pre - annotate the data. By moving the pre - annotation process from the backend to the front - end for processing, it greatly reduces the data transmission volume and the performance requirements for the server, shortens the time consumption, effectively avoids the problem of timeout failure, and improves the reliability.

[0007] To achieve the above - mentioned objectives, the present invention adopts the following technical solutions:

[0008] In the first aspect, the present invention provides a data annotation method based on WEB - end target recognition.

[0009] A data annotation method based on WEB - end target recognition includes the following processes:

[0010] After the image to be annotated is configured to the WEB - end, the WEB - end starts the pre - annotation function and directly calls the target detection model deployed in the WEB - end for pre - annotation to obtain the pre - annotation result;

[0011] Visualize the pre - annotation result in the image to be annotated on the WEB - end, count the usage frequencies of each label in the pre - annotation result and sort them in descending order for correcting the pre - annotation result. For labels with the same frequency, sort them in ascending order according to the first letter of the name.

[0012] In the second aspect, the present invention provides a data annotation system based on WEB - end target recognition.

[0013] A data annotation system based on WEB - end target recognition includes a WEB - end, where the target detection model is deployed in the WEB - end. In the WEB - end, there are configured:

[0014] A missing - label object adding unit for performing missing - label annotation operations on the WEB - end;

[0015] A mis - labeled object deleting unit for performing mis - labeled object deletion operations on the WEB - end;

[0016] A bounding - box boundary adjusting unit for performing bounding - box boundary adjustment operations on the WEB - end;

[0017] A labeled - object label modifying unit for performing labeled - object label modification operations on the WEB - end;

[0018] A labeled - object shape modifying unit for performing labeled - object shape modification operations on the WEB - end.

[0019] Compared with the prior art, the beneficial effects of the present invention are:

[0020] 1. The present invention adopts the method of calling the pre-annotation algorithm model on the WEB side to deploy the target detection model to the local WEB side. The WEB side directly calls the target detection model, and uses the computing power of the local computing card to pre-annotate the data through the WEB side. The pre-annotation process is advanced from the backend to the WEB side for processing, greatly reducing the data transmission volume and the performance requirements for the server, shortening the time-consuming, effectively avoiding the problem of communication timeout failure in the interface call between the front and back ends of the platform, and improving the reliability.

[0021] 2. The present invention provides the function of modifying the labels and shapes of the pre-annotation results, streamlining the process of first deleting the original mislabeled objects and then adding the missed-labeled objects. The label order is arranged in descending order of usage frequency and updated in real time, reducing the manual search time and improving the annotation work efficiency.

[0022] 3. The present invention provides the function of automatically generating a workload statistical report. During the secondary annotation process, a report is automatically generated, and the number of times of adding missed-labeled objects, deleting mislabeled objects, adjusting the boundaries of the annotation boxes, modifying object labels, and modifying object shapes is statistically counted in real time, saving the labor cost and time cost of manual statistics, being able to more conveniently and quickly evaluate the performance of the target detection model, providing reference for algorithm optimization engineers, and accelerating the iteration process of the target detection model.

[0023] Advantages of additional aspects of the present invention will be partly given in the following description, partly will become obvious from the following description, or will be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings forming a part of this specification are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0025] Figure 1 It is a schematic flowchart of the data annotation method provided in Embodiment 1 of the present invention;

[0026] Figure 2 It is a schematic flowchart of the training process of the target detection model provided in Embodiment 1 of the present invention;

[0027] Figure 3 It is a schematic diagram of the object with missed labels added provided in Embodiment 1 of the present invention;

[0028] Figure 4 It is a schematic diagram of the object with mislabels deleted provided in Embodiment 1 of the present invention;

[0029] Figure 5 It is a schematic diagram of adjusting the boundary of the annotation box provided in Embodiment 1 of the present invention;

[0030] Figure 6 It is a schematic diagram of modifying the label of the annotation object provided in Embodiment 1 of the present invention;

[0031] Figure 7 Schematic diagram of modifying the shape of the annotation object provided in Embodiment 1 of the present invention. Detailed implementation manners

[0032] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0033] It should be noted that the following detailed descriptions are all exemplary and are intended to provide further descriptions of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0034] In the case of no conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0035] Embodiment 1:

[0036] Embodiment 1 of the present invention provides a data annotation method based on target recognition on the WEB side, as Figure 1 shown, including the following processes:

[0037] S1: After the image to be annotated is configured to the WEB side, the WEB side starts the pre-annotation function and directly calls the target detection model deployed in the WEB side for pre-annotation to obtain the pre-annotation result;

[0038] S2: Visualize the pre-annotation result into the image to be annotated on the WEB side, count the usage frequency of each label in the pre-annotation result and sort them in descending order for correcting the pre-annotation result. For labels with the same frequency, sort them in ascending order according to the first letter of the name.

[0039] In this embodiment, the WEB (World Wide Web) is the global wide area network, also known as the World Wide Web. It is a global, dynamic interactive, cross-platform distributed graphic information system based on hypertext and HTTP. It is a network service built on the Internet and provides a graphical and easy-to-access intuitive interface for browsers to search and browse information on the Internet. The documents and hyperlinks therein organize the information nodes on the Internet into an interconnected network structure. The WEB side referred to in this embodiment refers to the web page side on a computer (or computer terminal). This embodiment combines the power transmission hidden danger annotation scenario and takes the TensorFlow.js call algorithm model as an example to elaborate in detail on the specific embodiments of the present invention.

[0040] In step S1, the deployment of the target detection model specifically includes:

[0041] On the WEB side, the trained object detection model is referenced. After the object detection model is trained, the object detection model is saved through the save_model method. The conversion tool TensorFlow.js is used to convert the object detection model so that the object detection model can be used in the browser. The model file model.json is directly referenced using the run method in the script, and the model is loaded using loadLayersModel. After setting the input, the pre-annotation results are generated for the input image.

[0042] In step S2, specifically, it includes:

[0043] Load the power transmission hidden danger image to be annotated as the input of the object detection model, call the object detection model for pre-annotation to obtain the pre-annotation results, visually display the pre-annotation results on the image to be annotated, and at the same time count the usage frequencies of each label in the pre-annotation results and sort them in descending order. If the frequencies are the same, sort them in ascending order by name.

[0044] In this embodiment, the correction of the pre-annotation results includes:

[0045] Manually view the pre-annotation results for secondary annotation. The present invention supports annotations of rectangles, polygons, points, polylines, and circles, and supports functions such as zooming in, zooming out, panning the image, and showing / hiding labels. The present invention provides five situations that require secondary annotation, including:

[0046] S2.1: Add objects that are missed in the annotation, such as a crane. As shown in, first select a shape close to the crane, such as a polygon, then click the mouse along the contour of the crane. Clicking once is for starting and pulling the line, and double-clicking is for ending. After double-clicking, connect the end point and the start point to form a closed polygon. A select label dialog box will pop up, and the labels are sorted in descending order of usage frequency. The higher the usage rate, the more forward it is, which is convenient for selection; Figure 3 After clicking to select the label DiaoChe and closing the select label dialog box, add the annotation result of the crane to the pre-annotation results, update the visual view display (that is, update the display of the image to be annotated on the WEB side), and it can be seen that the annotation of the crane is added to the view. The missed object crane is added successfully. The usage frequency of the selected label DiaoChe +1 (that is, the frequency increases by one), the labels are reordered, and at the same time, a record of the number of missed objects is added +1 in the workload statistics report;

[0047] After clicking to select the label DiaoChe and closing the select label dialog box, add the annotation result of the crane to the pre-annotation results, update the visual view display (that is, update the display of the image to be annotated on the WEB side), and it can be seen that the annotation of the crane is added to the view. The missed object crane is added successfully. The usage frequency of the selected label DiaoChe +1 (that is, the frequency increases by one), the labels are reordered, and at the same time, a record of the number of missed objects is added +1 in the workload statistics report;

[0048] S2.2: Delete mis-annotated objects, such as a crane. As shown in, click to select the mis-annotated crane object, which is highlighted in the visual view, and click the delete operation button or the keyboard back shortcut key to delete the annotation result of the mis-annotated object from the pre-annotation results; Figure 4 After clicking to select the mis-annotated crane object, which is highlighted in the visual view, and click the delete operation button or the keyboard back shortcut key to delete the annotation result of the mis-annotated object from the pre-annotation results;

[0049] Update the visualization view display. It can be seen that the mislabeled object crane has been deleted. The deletion of the mislabeled object crane is completed. The usage frequency of the mislabeled object label DiaoChe is decreased by 1 (i.e., the frequency is reduced by one time). The labels are reordered. At the same time, the record of the number of times the mislabeled object is deleted in the workload statistics report is incremented by 1;

[0050] S2.3: Adjust the boundary of the annotation box such as the crane. As Figure 5 shown, click to select the crane object whose annotation box boundary needs to be adjusted. It will be highlighted in the visualization view. Drag the boundary points of the annotation box with the mouse to adjust the boundary of the crane annotation box, and update the position of the annotation box in the visualization view in real time. Click on the position outside the annotation box boundary to cancel the selection state of the object. The adjustment of the boundary of the crane annotation box is completed. The record of the number of times the boundary of the annotation box is adjusted in the workload statistics report is incremented by 1;

[0051] S2.4: Modify the label of the annotation object. For example, change DiaoChe to TaDiao. As Figure 6 shown, click to select the label DiaoChe of the annotation object to be modified. A label selection dialog box will pop up. The labels are sorted in descending order of usage frequency. After clicking to select the label TaDiao, close the label selection dialog box, and change the label DiaoChe of the annotation result of the object to TaDiao;

[0052] Update the visualization view display. The label of the object is displayed as TaDiao. The operation of modifying the label of the annotation object is completed. The usage frequency of the original label DiaoChe of the annotation object is decreased by 1, and the usage frequency of the selected label TaDiao is increased by 1. The labels are reordered. At the same time, the record of the number of times the label of the annotation object is modified in the workload statistics report is incremented by 1;

[0053] S2.5: Modify the shape of the annotation object such as the tower crane. As Figure 7 shown, click to select the tower crane annotation object to be modified. It will be highlighted in the visualization view. Click on other annotation shape buttons or the corresponding shortcut keys to change the shape of the annotation result of the tower crane object to the newly selected shape;

[0054] Specifically, there are two cases. One is to add boundary points. For example, change from a rectangle to a polygon (the polygon in this embodiment specifically refers to other polygons excluding rectangles). Click the mouse on the boundary line, and a boundary point will be added at the midpoint of the boundary line. The other is to delete boundary points. For example, change from a polygon to a rectangle. The boundary points closest to the center point of the polygon will be deleted in sequence until there are 4 boundary points left. Then adjust the boundary of the annotation box to an appropriate position. Click on the position outside the annotation box boundary to cancel the selection state of the object. The modification of the shape of the annotation object is completed. The record of the number of times the shape of the annotation object is modified in the workload statistics report is incremented by 1. An example of the workload statistics report is shown in Table 1.

[0055] Table 1: Workload Statistical Report.

[0056]

[0057]

[0058] The workload report after secondary annotation in the present invention is used to evaluate the performance of the object detection model. As Figure 2 shown, it evaluates the reliability of the data annotation results of the object detection model, and uses the results of secondary annotation to continue adding to the training set to train the object detection model, realizing the iterative training of the object detection model.

[0059] Example 2:

[0060] Embodiment 2 of the present invention provides a data annotation system for object recognition based on the WEB side, including the WEB side, where the object detection model is deployed in the WEB side. In the WEB side, there are configured:

[0061] A missing label object adding unit for performing missing label annotation operations on the WEB side;

[0062] A mislabeled object deleting unit for performing mislabeled object deletion operations on the WEB side;

[0063] A bounding box boundary adjusting unit for performing bounding box boundary adjustment operations on the WEB side;

[0064] A labeled object label modifying unit for performing labeled object label modification operations on the WEB side;

[0065] A labeled object shape modifying unit for performing labeled object shape modification operations on the WEB side.

[0066] The missing label object adding unit is specifically configured as:

[0067] A shape similar to the target to be labeled is selected and labeled along the contour of the target to be labeled. When the mouse is clicked, it starts and draws a line, and when the mouse is double-clicked, the labeling ends. After the mouse is double-clicked, the end point and the start point are connected to form a closed labeling shape, and a label selection dialog box pops up, with the labels sorted in descending order of usage frequency;

[0068] After the label is selected, the label selection dialog box is closed, the annotation result is added to the pre-annotation result, the image to be labeled on the WEB side is updated, the usage frequency of the selected label is increased by one, the labels are reordered, and the number of missing label objects in the workload statistical report is increased by one. For a detailed example, refer to step S2.1 in Embodiment 1.

[0069] The mislabeled object deleting unit is specifically configured as:

[0070] When the mislabeled target object is selected, it is highlighted in the image to be labeled on the WEB side, the mislabeled target object is deleted from the pre-labeling result, the usage frequency of the label of the mislabeled object is reduced by one, the labels are reordered, and the number of mislabeled objects in the workload statistics report is increased by one. For a detailed example, see step S2.2 in Embodiment 1.

[0071] The annotation box boundary adjustment unit is specifically configured as follows:

[0072] After the target object whose annotation box boundary needs to be adjusted is selected, it is highlighted in the image to be labeled on the WEB side, the boundary point positions of the annotation box are dragged and moved by the mouse, and the boundary of the annotation box of the target object is adjusted;

[0073] The position of the annotation box in the image to be labeled on the WEB side is updated in real time. When a position outside the boundary of the annotation box of the target object is clicked, the selection state of the target object is cancelled, and the number of times of adjusting the annotation box boundary in the workload statistics report is increased by one. For a detailed example, see step S2.3 in Embodiment 1.

[0074] The annotation object label modification unit is specifically configured as follows:

[0075] When the first label to be modified is selected, a label selection dialog box pops up, the labels are sorted in descending order of usage frequency, and after the second label is clicked and selected, the label selection dialog box closes, and the second label replaces the first label;

[0076] The replacement result is updated to the image to be labeled on the WEB side, the usage frequency of the first label is reduced by one, the usage frequency of the second label is increased by one, the labels are reordered, and the number of times of modifying the label of the annotation object in the workload statistics report is increased by one. For a detailed example, see step S2.4 in Embodiment 1.

[0077] The annotation object shape modification unit is specifically configured as follows:

[0078] When the annotation shape to be modified is selected, it is highlighted in the image to be labeled on the WEB side, and other annotation shape buttons or corresponding shortcut keys are clicked and selected, and the annotation shape of the annotation object is changed to the selected shape;

[0079] When a boundary point needs to be added, when the mouse is clicked on the boundary line, a boundary point is added at the midpoint of the boundary line; when a boundary point needs to be deleted, the boundary point closest to the center point of the polygon is deleted in sequence until the remaining set number of boundary points (for example, when modifying the polygon to a rectangle, it is until 4 boundary points remain, and when modifying to a line, it is until 2 boundary points remain. Here, it will not be enumerated one by one, and the number of boundary points can be selected according to the shape to be modified).

[0080] Adjust the boundary of the current annotation box to a suitable position. When a position outside the boundary of the annotation box is clicked, the selected state of the current annotation box is cancelled, and the number of times of modifying the shape of the annotation object in the workload statistics report is increased by one. For a detailed example, refer to step S2.5 in Embodiment 1.

[0081] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A data annotation method based on web - end target recognition, characterized in that, The process includes the following: After the image to be labeled is configured on the WEB side, the WEB side starts the pre-labeling function, directly calls the target detection model deployed on the WEB side for pre-labeling, and obtains the pre-labeling result; The pre-labeling result is visually displayed in the image to be labeled on the WEB side, and the usage frequency of each label in the pre-labeling result is counted and sorted in descending order for correcting the pre-labeling result. For labels with the same frequency, they are sorted in ascending order according to the first letter of the name.

2. The data annotation method based on web - end target recognition according to claim 1, characterized in that, The labeling types during the correction of the pre-labeling result include at least: rectangle labeling, polygon labeling, point labeling, polyline labeling, and circular labeling; The labeling methods during the correction of the pre-labeling result include at least: label magnification, label reduction, image translation, label display, and label hiding; The magnification and reduction of the label depend on the magnification and reduction of the image. When the mouse wheel is scrolled, the image is magnified or reduced, and the labeled labels and labeling frames on the image are magnified or reduced accordingly.

3. The data annotation method based on web - end target recognition according to claim 1, characterized in that, After the pre-labeling result is visually displayed in the image to be labeled on the WEB side, the correction of the labeling result is also carried out, including: The shape similar to the object to be labeled is selected, and the contour of the object to be labeled is labeled. Clicking the mouse is the start and drawing of the line, and double-clicking the mouse is the end of the labeling. After double-clicking the mouse, the end point and the start point are connected to form a closed labeling shape, and the label selection dialog box pops up. The labels are sorted in descending order of usage frequency; After the label is selected, the label selection dialog box is closed, the labeling result is added to the pre-labeling result, the image to be labeled on the WEB side is updated, the usage frequency of the selected label is increased by one, the labels are re-sorted, and the number of missed-labeled objects in the workload statistics report is increased by one.

4. The data annotation method based on web - end target recognition according to claim 1, characterized in that, After the pre-labeling result is visually displayed in the image to be labeled on the WEB side, the correction of the labeling result is also carried out, including: When the mis-labeled target object is selected, it is highlighted in the image to be labeled on the WEB side, the mis-labeled target object is deleted from the pre-labeling result, the usage frequency of the label of this mis-labeled object is reduced by one, the labels are re-sorted, and the number of mis-labeled objects in the workload statistics report is increased by one.

5. The data annotation method based on web - end target recognition according to claim 1, characterized in that, After the pre-labeling result is visually displayed in the image to be labeled on the WEB side, the correction of the labeling result is also carried out, including: After the target object whose labeling frame boundary needs to be adjusted is selected, it is highlighted in the image to be labeled on the WEB side, and the position of the boundary point of the labeling frame is dragged and moved by the mouse, and the boundary of the labeling frame of the target object is adjusted; The position of the labeling frame in the image to be labeled on the WEB side is updated in real time. When a position outside the boundary of the labeling frame of the target object is clicked, the selected state of the target object is cancelled, and the number of times of adjusting the labeling frame boundary in the workload statistics report is increased by one.

6. The data annotation method based on web - end target recognition according to claim 1, characterized in that, After the pre-labeling result is visually displayed in the image to be labeled on the WEB side, the correction of the labeling result is also carried out, including: The first label to be modified is selected, the label selection dialog box pops up, the labels are sorted in descending order of usage frequency. After the second label is clicked and selected, the label selection dialog box is closed, and the second label replaces the first label; Update the replacement result to the image to be labeled on the WEB side. The usage frequency of the first label is decreased by one, the usage frequency of the second label is increased by one, the labels are reordered, and the number of times the label of the labeled object is modified in the workload statistics report is increased by one.

7. The data annotation method based on web - end target recognition according to claim 1, characterized in that, After visualizing the pre-labeling result on the image to be labeled on the WEB side, perform correction of the labeling result, including: The labeling shape to be modified is selected and highlighted in the image to be labeled on the WEB side. Other labeling shape buttons or corresponding shortcut keys are clicked and selected, and the labeling shape of the labeled object is changed to the selected shape; When adding boundary points is required, when the mouse is clicked on the boundary line, a boundary point is added at the midpoint of the boundary line. When deleting boundary points is required, the boundary points closest to the center point of the polygon are deleted in sequence until the number of boundary points remaining matches the target graphic to be modified; Adjust the boundary of the current labeling box to a suitable position. When a position outside the boundary of the labeling box is clicked, the selection state of the current labeling box is cancelled, and the number of times the shape of the labeled object is modified in the workload statistics report is increased by one.

8. The data annotation method based on web - end target recognition according to any one of claims 1 - 7, characterized in that, Perform performance evaluation of the target detection model according to the workload statistics report, and correct the target detection model in combination with the result of the pre-labeling result until the pre-labeling result is consistent with the manual marking result.

9. A data annotation system based on web - end target recognition, characterized in that, Including the WEB side, the target detection model is deployed in the WEB side. In the WEB side, there are configured: A missing label object adding unit for performing missing label marking operations on the WEB side; A mislabeled object deleting unit for performing mislabeled object deletion operations on the WEB side; A labeling box boundary adjustment unit for performing labeling box boundary adjustment operations on the WEB side; A labeled object label modification unit for performing labeled object label modification operations on the WEB side; A labeled object shape modification unit for performing labeled object shape modification operations on the WEB side.

10. The data annotation system for target recognition based on the WEB side according to claim 9, characterized in that, The missing label object adding unit is configured as: A shape similar to the object to be labeled is selected and labeled along the contour of the object to be labeled. When the mouse is clicked once, it is for starting and pulling the line, and when the mouse is double-clicked, it is for ending the labeling. After the mouse is double-clicked, the end point and the start point are connected to form a closed labeling shape, and a label selection dialog box pops up. The labels are sorted in descending order of usage frequency; After the label is selected, the label selection dialog box is closed, the labeling result is added to the pre-labeling result, the image to be labeled on the WEB side is updated, the usage frequency of the selected label is increased by one, the labels are reordered, and the number of missing label objects in the workload statistics report is increased by one; Or, The mislabeled object deleting unit is configured as: When the mislabeled target object is selected, it is highlighted in the image to be labeled on the WEB side. The mislabeled target object is deleted from the pre-labeling result. The usage frequency of the label of this mislabeled object is decreased by one, the labels are reordered, and the number of mislabeled objects in the workload statistics report is increased by one; Or, The labeling box boundary adjustment unit is configured as: After the target object whose labeling box boundary needs to be adjusted is selected, it is highlighted in the image to be labeled on the WEB side. The boundary point positions of the labeling box are dragged and moved by the mouse, and the boundary of the labeling box of the target object is adjusted; The position of the annotation box in the image to be annotated on the WEB side is updated in real time. When a position outside the boundary of the annotation box of the target object is clicked, the selected state of the target object is cancelled, and the number of times of adjusting the boundary of the annotation box in the workload statistics report is increased by one time. Or, The annotation object label modification unit is configured as follows: When the first label to be modified is selected, a label selection dialog box pops up. The labels are sorted in descending order of usage frequency. After the second label is clicked and selected, the label selection dialog box closes, and the second label replaces the first label; The replacement result is updated to the image to be annotated on the WEB side. The usage frequency of the first label is decreased by one time, the usage frequency of the second label is increased by one time, the labels are reordered, and the number of times of modifying the annotation object label in the workload statistics report is increased by one time; Or, The annotation object shape modification unit is configured as follows: When the annotation shape to be modified is selected, it is highlighted in the image to be annotated on the WEB side. When other annotation shape buttons or corresponding shortcut keys are clicked and selected, the annotation shape of the annotation object is changed to the selected shape; When a boundary point needs to be added, when the mouse is clicked on the boundary line, a boundary point is added at the midpoint of the boundary line; When a boundary point needs to be deleted, the boundary point closest to the center point of the polygon is deleted in sequence until the number of boundary points remaining matches the target graphic to be modified; The boundary of the current annotation box is adjusted to a suitable position. When a position outside the boundary of the annotation box is clicked, the selected state of the current annotation box is cancelled, and the number of times of modifying the annotation object shape in the workload statistics report is increased by one time.

Images