A method, system, and computer-readable storage medium for fusion of annotation categories

Abstract

This application belongs to the field of data annotation and discloses a method, system, and computer-readable storage medium for fusion of annotation categories. The method includes: obtaining basic annotation category text and incremental annotation category text, and maintaining full annotation category text; traversing the basic annotation category text and the incremental annotation category text, and fusing the annotation content nodes of the basic annotation category text and the incremental annotation category text with reference to the node connection method of the full annotation category text to form the annotation content nodes and connection method of the fused annotation category text, wherein the connection method of the formal nodes of the full annotation category text is recorded using the annotation framework of the full annotation category text; deleting redundant stages in the annotation framework of the full annotation category text according to the fused annotation category text to form a new annotation framework, and merging the fused annotation category text and the new annotation framework to form the fused annotation category text, which can achieve the effect of flexibly deleting and modifying the decision graph.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data annotation, in particular to an annotation category fusion method and system and a computer readable storage medium. BACKGROUND

[0002] With the increase of collected data and annotation categories, there is an increasing demand for supplementing annotation of new categories or modifying annotation of existing categories to adapt to the iteration of deep learning models. This will cause difficulties in managing annotated content.

[0003] The current annotation platform can use a decision graph to represent annotation categories, the dependency relationship between categories (such as the car door frame being an attached frame of the car frame, so it depends on the car frame), and the required front-end interaction (staging annotation categories, and a class of front-end interaction is required for categories in a stage), and a simple decision graph fusion strategy is used to meet the supplementing annotation demand. The existing annotation platform uses a decision graph to annotate categories. When supplementing annotation of annotation text is required, due to the flexible and variable demand for supplementing annotation, technical personnel need to modify the code of the annotation platform to add corresponding supplementing annotation functions to meet the demand for supplementing annotation. However, the code of the annotation platform is huge, and the coupling and correlation between functions are strong. When technical personnel modify the code of the annotation platform to add supplementing annotation functions, it is difficult to locate the code that needs to be added to supplement the annotation function, and the logic of code modification is relatively complex. The existing annotation platform is difficult to adapt to flexible supplementing annotation scenarios. SUMMARY

[0004] To this end, embodiments of the present application provide an annotation category fusion method, system and computer readable storage medium, which can solve the technical problem that it is difficult to adapt to flexible supplementing annotation scenarios when using a decision graph to annotate categories. The specific technical solution content is as follows:

[0005] In a first aspect, embodiments of the present application provide an annotation category fusion method, which comprises:

[0006] obtaining a basic annotation category text and an incremental annotation category text, and maintaining a full-amount annotation category text; wherein the basic annotation category text, the incremental annotation category text and the full-amount annotation category text are provided with the same node type, structure and annotation content expression form, the node type includes an annotation content node and a form node, and the full-amount annotation category text contains annotation content nodes of all supported annotation types within a business scope and all form nodes;

[0007] traverse the basic annotation category text and the incremental annotation category text, refer to the node connection mode of the full-amount annotation category text, fuse the annotation content nodes of the basic annotation category text and the incremental annotation category text, form the annotation content nodes and the connection mode of the fusion annotation category text, wherein the connection mode of the form node of the full-amount annotation category text is recorded in the annotation framework of the full-amount annotation category text;

[0008] delete the redundant stages in the annotation framework of the full-amount annotation category text according to the fusion annotation category text, form a new annotation framework, combine the fusion annotation category text and the new annotation framework, and form a fusion annotation category text.

[0009] Optionally, the types of the form nodes include task start nodes and task end nodes, stage start nodes and stage end nodes, and cluster nodes; the task start nodes and the task end nodes are only one in the CTG annotation category text respectively, and the task start nodes and the task end nodes appear in pairs; the stage start nodes and the stage end nodes are used to define front-end interaction operation types, and the stage start nodes and the stage end nodes appear in pairs; the cluster nodes represent the dependency relationship between annotation categories; and the annotation content nodes represent annotation categories.

[0010] Optionally, the traversing the basic annotation category text and the incremental annotation category text, referring to the node connection mode of the full-amount annotation category text, fusing the annotation content nodes of the basic annotation category text and the incremental annotation category text, and forming the annotation content nodes and the connection mode of the fusion annotation category text comprise:

[0011] checking each annotation content node in the basic annotation category text and the incremental annotation category text, referring to the full-amount annotation category text, moving upwards or downwards in the full-amount annotation category text until the first stage start node and the stage end node of the full-amount annotation category text are encountered, and adding the nodes found in the full-amount annotation category text and lines corresponding to the nodes to the fusion annotation category text.

[0012] Optionally, the deleting the redundant stages in the annotation framework of the full-amount annotation category text according to the fusion annotation category text, and forming a new annotation framework comprise:

[0013] specifying the parallel stages in the annotation framework of the fusion annotation category text and the full-amount annotation category text through the task start nodes, deleting the redundant stages in the annotation framework of the full-amount annotation category text according to the fusion annotation category text, and forming a new annotation framework, wherein the stages are formed by pairs of the stage start nodes and the stage end nodes.

[0014] Optionally, the deleting the redundant stage in the annotation framework according to the fusion annotation category text comprises:

[0015] Comparing the annotation framework of the fusion annotation category text with the annotation framework of the full-amount annotation category text, if there is a stage existing in the annotation framework of the full-amount annotation category text but not existing in the fusion annotation category text, if the end point of any line of the annotation framework of the full-amount annotation category text points to the stage start node of the stage, the end point of the line is pointed to the stage start node of the next stage of the stage, and other lines starting from the stage end node are deleted.

[0016] Optionally, the method further comprises:

[0017] Starting from the task start node, the DSF algorithm is used to traverse the full-amount annotation category text to form the annotation framework of the full-amount annotation category text.

[0018] Optionally, the method further comprises:

[0019] If the annotation category needs to be modified, when the annotation content nodes of the fusion basic annotation category text and the incremental annotation category text are formed, the old annotation category in the basic annotation category text is ignored.

[0020] Optionally, the node is set as (box / property_id, node_id, meta), the box / property_id is an attribute label, the node_id is a node label, and the meta is a node definition annotation content format.

[0021] In a second aspect, an embodiment of the present application provides an annotation category fusion system, the system comprises:

[0022] The processing module is configured to obtain a basic annotation category text and an incremental annotation category text, and maintain a full-amount annotation category text; wherein the basic annotation category text, the incremental annotation category text and the full-amount annotation category text are provided with the same node type, structure and annotation content expression form, the node type comprises an annotation content node and a form node, and the full-amount annotation category text comprises annotation content nodes of all supported annotation types in a business range and all form nodes; the basic annotation category text and the incremental annotation category text are traversed, the annotation content nodes of the basic annotation category text and the incremental annotation category text are fused in the node connection mode of the full-amount annotation category text as a reference, to form annotation content nodes and connection modes of a fused annotation category text, and the connection mode of the form nodes of the full-amount annotation category text is recorded in an annotation framework of the full-amount annotation category text;

[0023] The second fusion module is configured to delete redundant stages in the annotation framework of the full-amount annotation category text according to the fused annotation category text, form a new annotation framework, combine the fused annotation category text and the new annotation framework, and form a fused annotation category text.

[0024] In a third aspect, an embodiment of the present application provides a computer readable storage medium storing a computer program, and the computer program is executed by a processor to implement the steps of the annotation category fusion method in any of the preceding embodiments.

[0025] In summary, compared with the prior art, the technical scheme provided by the embodiments of the present application has at least the following beneficial effects:

[0026] By standardizing the node type and structure of the annotation category text and the expression form of the annotation content, the basic annotation category text and the incremental annotation category text can be standardized and compared in each stage when the annotation category text fusion or comparison is performed, the forest is converted into a tree, the unified algorithm can be used to solve the problem of the tree and the forest when the annotation category text fusion is performed, the full-amount annotation category text is taken as a reference for the fusion of the basic annotation category text and the incremental annotation category text, the annotation framework of the full-amount annotation category text is modified into a new annotation framework according to the fused annotation category text, and the fused annotation category text is combined with the new annotation framework, so that the fused annotation category text is legal in the new annotation framework, the problem that the existing decision graph setting and fusion strategy do not support the insertion stage and the deletion and modification of the node and cannot meet the demand of the new front-end interaction and the modification of the annotation specification is solved, and when the new interaction mode or the modification of the annotation specification is added in the front end, the legal decision graph cannot be formed. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1is a flowchart of a labeling category fusion method provided by one embodiment of the present application.

[0028] Figure 2 is one of flowcharts of a labeling category fusion method provided by another embodiment of the present application.

[0029] Figure 3 is another of flowcharts of a labeling category fusion method provided by another embodiment of the present application. DETAILED DESCRIPTION

[0030] The specific embodiments are merely illustrative of the present application, and are not intended to limit the present application. Those skilled in the art can make modifications to the embodiments without creative contribution, and the modifications are within the scope of the present application as long as they are within the scope of the claims of the present application.

[0031] To make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative contribution are within the scope of protection of the present application.

[0032] In addition, the term “and / or” in the present application is merely to describe the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can mean that there are three cases of A alone, A and B together, and B alone. In addition, the character “ / ” in the present application generally represents an “or” relationship between the front and rear associated objects unless otherwise specified.

[0033] The terms “first”, “second”, and the like in the present application are used to distinguish the same items or similar items with basically the same function and action, and it should be understood that there is no logical or time sequence relationship between “first”, “second”, and “n”, and the quantity and execution order are not limited.

[0034] UnionTask is a general labeling platform across labeling tools, which can perform image labeling, image segmentation, point cloud labeling and the like. UnionTask provides general data import, labeling, re-labeling, settlement processes, and any labeling tool connected to UnionTask can use the process services provided by UnionTask.

[0035] The current annotation platform Union Task can represent the annotation categories and the dependency relationship between the categories and the required front-end interaction by a decision graph. For example, the door frame is an auxiliary frame of the car frame, so it depends on the car frame. The annotation categories are divided into stages, the categories in a stage require a type of front-end interaction, and a simple decision graph fusion strategy is used to meet the re-labeling requirements. In the re-labeling process, Union Task defines the concept of annotation category text (Category) to describe the content that needs to be annotated for a piece of data. The annotation category text is a string form of the decision graph, which is divided into Node, Edge, and Option three parts, and can completely describe a visual graph (decision graph). The decision graph is based on the decision graph, but further defines different types of nodes and stages,

[0036] and the described objects are 'annotation content' and 'front-end interaction mode during annotation'.

[0037] The embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

[0038] Reference Figure 1 In an embodiment of the present application, an annotation category fusion method is provided, and the main steps of the method are described as follows:

[0039] S1: Obtain the basic annotation category text and the incremental annotation category text, and maintain the full-amount annotation category text; wherein the basic annotation category text, the incremental annotation category text, and the full-amount annotation category text are provided with the same node type, structure, and annotation content expression form, the node type includes an annotation content node and a form node, and the full-amount annotation category text contains annotation content nodes of all supported annotation types in the business scope and all form nodes;

[0040] S2: Traverse the basic annotation category text and the incremental annotation category text, and fuse the annotation content nodes of the basic annotation category text and the incremental annotation category text in the node connection mode of the full-amount annotation category text to form the annotation content nodes and the connection mode of the fusion annotation category text, wherein the connection mode of the form nodes of the full-amount annotation category text is recorded in the annotation framework of the full-amount annotation category text;

[0041] S3: Delete the redundant stages in the annotation framework of the full-amount annotation category text according to the fusion annotation category text to form a new annotation framework, combine the fusion annotation category text and the new annotation framework to form a fusion annotation category text.

[0042] Specifically, in the embodiment, the node type and structure specification in the annotation category text are specified, i.e., the fixed type required for dividing the nodes in the annotation category text, and the corresponding structure of each type of node is set. In the embodiment, the node types include an annotation content node and a form node. The annotation content node contains the specific content of the current annotation, and the form node respectively describes each node of the start of the annotation task, the dependency relationship between the annotation categories, and the annotation stage. One annotation stage represents a type of front-end interaction mode required by the category. For example, the nodes representing the annotation stage include an RIA frame start node and an RIA attribute start node. The RIA frame start node means that the front end switches to the frame interaction mode (selection box type + mouse drag frame), and the annotator can select the pedestrian or car type under the 2D or 3D frame on the front end. After completing the frame, the front end switches to the attribute interaction mode (selected box + attribute) in the RIA attribute start node. If the car frame (corresponding to cluster 1) is selected, the Left Turn attribute and the Right Turn attribute need to be annotated. If the pedestrian frame (corresponding to cluster 2) is selected, the Ped_Occ attribute needs to be annotated.

[0043] In the embodiment, a full-amount annotation category text is maintained. The full-amount annotation category text includes the annotation content nodes and all form nodes of all supported annotation types in the business scope. In data annotation, a full-amount annotation category text also needs to be maintained. Therefore, the full-amount annotation category text required to be maintained in the application does not require additional maintenance cost. There is only one full-amount annotation category text, and different annotation scenarios can be represented by different annotation category texts. For example, annotation category text 1 annotates people and cars, corresponding to one annotation scenario, annotation category text 2 annotates people and trees, corresponding to another scenario, and the full-amount annotation category text includes people, cars, and trees.

[0044] Specifically, the form node includes a stage start node and a stage end node, and the stage start node and the stage end node appear in pairs. The DSF algorithm is used to traverse the full-amount annotation category text to form an annotation framework of the full-amount annotation category text. The annotation framework of the full-amount annotation category text represents the connection mode of the form nodes in the full-amount annotation category text, such as the order of the stage start / ending nodes, and does not include the annotation content nodes. The specification stipulates that the stage start / ending nodes and the stage end nodes appear in pairs, and the DSF algorithm is used to make the start / ending nodes of the same stage appear adjacent in the access order, so as to facilitate the extraction of the annotation framework. Further, a stage is expressed by the paired stage start node and the stage end node. In the embodiment, the annotation framework of the full-amount annotation category text represents the relationship between the stages.

[0045] Specifically, when forming the fused annotation category text, the full annotation category text is used as a reference framework. Referring to the node position information within the full annotation category text, the basic annotation category text and the incremental annotation category text are merged to form the fused annotation category text. This integrates scattered supplementary annotation content into a unified annotation text framework, transforming multiple supplementary annotation tasks from a forest into a tree. In this embodiment, the basic annotation category text refers to the already annotated annotation categories, the fused annotation category text refers to the categories of interest to the supplementary annotation requirements, and the fused annotation category text refers to the annotation categories included in the generated supplementary annotation tasks. Furthermore, in the fused annotation category text, annotation content nodes are connected to the stage start node or stage end node.

[0046] Specifically, in this embodiment, redundant stages in the annotation framework of the full annotation category text are deleted according to the fusion annotation category text. That is, the annotation frameworks of the fusion annotation category text and the full annotation category text are compared. If there are stages in the annotation framework of the full annotation category text that are not in the fusion annotation category text, then the stage is deleted in the annotation framework of the full annotation category text. The modified annotation framework of the full annotation category text is then used to form a new annotation framework. The new annotation framework is then merged with the fusion annotation category text to form a valid fusion annotation category text.

[0047] This application standardizes the node types and structure of labeled category texts and the expression of labeled content. This allows for the parallel arrangement of standards at each stage of labeled category text during fusion or comparison, transforming a forest into a tree structure. A unified algorithm can then be used to address the tree-forest problem during labeled category text fusion. The entire set of labeled category texts serves as the base and reference for fusion of incremental labeled category texts. The labeling framework of the entire set of labeled category texts is modified to a new labeling framework based on the fused labeled category text. This new framework is then used to merge the fused labeled category texts, ensuring their validity. This addresses the problem that existing decision graph settings and fusion strategies do not support inserting stages or deleting / modifying nodes, failing to meet the needs of adding new front-end interactions and modifying labeling standards. Furthermore, it prevents the formation of a valid decision graph when adding new interaction methods or modifying labeling standards on the front end.

[0048] Furthermore, in another embodiment, the node types and structures of the labeled category text are set to the same type and expression.

[0049] The types of format nodes include task start nodes and task end nodes, phase start nodes and phase end nodes, and cluster nodes; wherein, there is only one task start node and one task end node in the labeled category text, and task start nodes and task end nodes appear in pairs; both phase start nodes and phase end nodes are used to define the front-end interactive operation type, and phase start nodes and phase end nodes appear in pairs; the cluster node represents the dependency relationship between labeled categories; the labeled content node represents the labeled category.

[0050] Specifically, in this implementation, the node type is defined by the Node_Type field. The task start node and the task end point have no labeling meaning, but the labeling category text structure can be standardized into a tree without losing expressive power, thus eliminating the forest.

[0051] Furthermore, the specification for the annotation category text pairs of nodes is (box / property_id, node_id, meta), where box / property_id is the attribute label, node_id is the node label, and meta is the annotation content format defined by the node. (box / property_id, node_id, meta) indicates which annotation category text node a specific annotation result belongs to, and meta is the annotation content format defined by each node.

[0052] Reference Figure 2 Optionally, in another embodiment, S2 is S2':

[0053] S2': Check each annotation content node in the basic annotation category text and the incremental annotation category text. Using the full annotation category text as a reference, move up or down in the full annotation category text until the first stage start node and the stage end node of the full annotation category text are encountered. Add the node found in the full annotation category text and the line corresponding to the node to the fused annotation category text.

[0054] Specifically, in this embodiment, a Graph Walk is performed by moving up or down within the fully labeled category text frame.

[0055] Reference Figure 3 Optionally, in another embodiment, S3 includes:

[0056] S31: By standardizing the parallel arrangement of stages in the annotation framework of the fused annotation category text and the full annotation category text through the task start node, and deleting redundant stages in the annotation framework of the full annotation category text according to the fused annotation category text to form a new annotation framework, wherein the stage is formed by a pair of stage start nodes and stage end nodes.

[0057] S32: Merge the fused annotation category text and the new annotation framework to form the fused annotation category text.

[0058] Specifically, in this embodiment, by corresponding the task start node, the stages in the annotation framework of the fused annotation category text and the full annotation category text are made to correspond, transforming the forest problem into a tree problem. This facilitates the subsequent comparison of the annotation frameworks of the fused annotation category text and the full annotation category text, identifies the stages in the annotation framework of the full annotation category text that are more numerous than those in the fused annotation category text, and deletes these stages so that the stages in the annotation framework of the full annotation category text correspond to those in the fused annotation category text.

[0059] Furthermore, in another embodiment, deleting redundant stages in the annotation framework based on the fused annotation category text is as follows: compare the annotation framework of the fused annotation category text with that of the full annotation category text. If there is a stage that exists in the annotation framework of the full annotation category text but not in the fused annotation category text, and if the endpoint of any line in the annotation framework of the full annotation category text points to the stage start node of the stage, then point the endpoint of the line to the stage start node of the next stage of the stage, and delete other lines originating from the stage end node of the stage.

[0060] Optionally, in another implementation, if it is necessary to modify the annotation category, when merging the annotation content nodes of the basic annotation category text and the incremental annotation category text to form the annotation content nodes and connection method of the merged annotation category text, the old annotation category in the basic annotation category text is ignored.

[0061] Specifically, the settings in this application allow for the addition of new stages to incremental annotation category text and even changes to the stages of annotation category text. This satisfies the need to add new interaction types and modify the annotation specifications of existing annotation categories. For example, if the annotation specifications change, a correction stage can be added after all annotation category texts, and the corresponding interaction can be added to the front end. After all tasks are annotated, the correction stage of the composite annotation category text for this batch of tasks can be deleted uniformly.

[0062] If you need to modify the label category (for example, change the traffic cone category to the stationary obstacle category - traffic cone attribute), you need to configure the basic label category text Projection List to express the old category to be transformed and the new category after transformation. In step 3 of merging label category text, ignore the old label category of the basic label category text to synthesize a valid merged label category text.

[0063] When generating supplementary annotation tasks based on basic tasks, thanks to the previous standardization of annotation content format, when the annotation content of a new node encounters the annotation information of an old node, the answer of that node can be formally transformed into the answer of the new node. For example: {Box_id:1,Node:Traffic cone}→{Box_id:1,Node:Stationary obstacle}+{Property_id=1,option=Traffic cone,meta:{parent_box_id:1}}.

[0064] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0065] In one embodiment of this application, a label category fusion system is provided, which corresponds one-to-one with the label category fusion methods in the above embodiments. The label category fusion system includes:

[0066] The processing module is used to acquire basic annotation category text and incremental annotation category text, and maintain full annotation category text. The basic annotation category text, incremental annotation category text, and full annotation category text all have the same node type, structure, and annotation content expression form. The node type includes annotation content nodes and form nodes. The full annotation category text includes annotation content nodes and all form nodes for all supported annotation types within the business scope. The first fusion module is used to traverse the basic annotation category text and the incremental annotation category text, and, with reference to the node connection method of the full annotation category text, fuse the annotation content nodes of the basic annotation category text and the incremental annotation category text to form the annotation content nodes and connection method of the fused annotation category text. The connection method of the form nodes of the full annotation category text is recorded using the annotation framework of the full annotation category text.

[0067] The second fusion module is used to delete redundant stages in the annotation framework of the full annotation category text according to the fusion annotation category text, form a new annotation framework, and merge the fusion annotation category text and the new annotation framework to form the fusion annotation category text.

[0068] Furthermore, in another embodiment, the types of formal nodes include task start nodes and task end nodes, phase start nodes and phase end nodes, and cluster nodes; wherein, there is only one task start node and one task end node in the labeled category text, and task start nodes and task end nodes appear in pairs; both phase start nodes and phase end nodes are used to define front-end interactive operation types, and phase start nodes and phase end nodes appear in pairs; the cluster node represents the dependency relationship between labeled categories; the labeled content node represents the labeled category.

[0069] Furthermore, in another embodiment, the step of traversing the basic annotation category text and the incremental annotation category text, and using the node connection method of the full annotation category text as a reference, merging the annotation content nodes of the basic annotation category text and the incremental annotation category text to form the annotation content nodes and connection method of the merged annotation category text, includes:

[0070] Examine each annotation content node in the basic annotation category text and the incremental annotation category text. Using the full annotation category text as a reference, move up or down in the full annotation category text until the first stage start node and the stage end node of the full annotation category text are encountered. Add the node found in the full annotation category text and the line corresponding to the node to the fused annotation category text.

[0071] Furthermore, in another embodiment, the step of deleting redundant stages in the annotation framework based on the fused annotation category text to form a new annotation framework includes:

[0072] By standardizing the parallel arrangement of stages in the annotation framework of the fused annotation category text and the full annotation category text through the task start node, redundant stages in the annotation framework of the full annotation category text are deleted according to the fused annotation category text to form a new annotation framework, wherein the stage is formed by a pair of stage start nodes and stage end nodes.

[0073] Furthermore, in another embodiment, the step of deleting redundant elements from the annotation framework based on the fused annotation category text includes:

[0074] Compare the annotation frames of the fused annotation category text and the full annotation category text. If there is a stage that exists in the annotation frame of the full annotation category text but not in the fused annotation category text, and if the endpoint of any line in the annotation frame of the full annotation category text points to the stage start node of the stage, then point the endpoint of the line to the stage start node of the next stage of the stage, and delete other lines originating from the stage end node of the stage.

[0075] Furthermore, in another embodiment, the method further includes:

[0076] Starting from the task initiation node, the DSF algorithm is used to traverse all labeled category texts to form a labeling framework for all labeled category texts. The labeling framework for all labeled category texts expresses the relationship between stages.

[0077] Furthermore, in another embodiment, the method further includes:

[0078] If it is necessary to modify the annotation category, when merging the annotation content nodes of the basic annotation category text and the incremental annotation category text to form the annotation content nodes and connection method of the merged annotation category text, the old annotation category in the basic annotation category text will be ignored.

[0079] Furthermore, in another embodiment, the node is set as (box / property_id, node_id, meta), where box / property_id is the attribute label, node_id is the node label, and meta is the annotation content format defined for the node.

[0080] The various modules of the aforementioned labeling category fusion system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0081] In one embodiment of this application, a computer-readable storage medium is provided, which stores a computer program. When executed by a processor, the computer program implements the labeling category fusion method steps described in the above embodiments. The computer-readable storage medium includes ROM (Read-Only Memory), RAM (Random-Access Memory), CD-ROM (Compact Disc Read-Only Memory), magnetic disk, floppy disk, etc.

[0082] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system described in this application can be divided into different functional units or modules to complete all or part of the functions described above.

Claims

1. A method for fusing annotated classes, the method comprising: The method comprises: acquiring a basic annotation category text and an incremental annotation category text, and maintaining a full-amount annotation category text; wherein the basic annotation category text, the incremental annotation category text and the full-amount annotation category text are provided with the same node type, structure and annotation content expression form, the node type comprises an annotation content node and a form node, the full-amount annotation category text comprises annotation content nodes of all supported annotation types within a business range and all form nodes, the node type and structure of the annotation category text are specified, i.e. the nodes in the annotation category text are divided into fixed types as required, and the nodes of each type are provided with corresponding structures, the form node type comprises a task start node and a task end node, a stage start node and a stage end node, and a clustering node, wherein the task start node and the task end node have no annotation meaning, but can specify the annotation category text structure as a tree without losing expression capability, and eliminate forests; traversing the basic annotation category text and the incremental annotation category text, taking the node connection mode of the full-amount annotation category text as a reference, fusing the annotation content nodes of the basic annotation category text and the incremental annotation category text, and forming the annotation content nodes and connection mode of the fused annotation category text, wherein the connection mode of the form nodes of the full-amount annotation category text is recorded in the annotation framework of the full-amount annotation category text; according to the fused annotation category text, deleting the redundant stages in the annotation framework of the full-amount annotation category text, forming a new annotation framework, merging the fused annotation category text and the new annotation framework, and forming a fused annotation category text.

2. The label category fusion method of claim 1, wherein, The form node type comprises a task start node and a task end node, a stage start node and a stage end node, and a clustering node; wherein the task start node and the task end node are only one in the annotation category text respectively, and the task start node and the task end node appear in pairs; the stage start node and the stage end node are used for defining front-end interaction operation types, and the stage start node and the stage end node appear in pairs; the clustering node represents a dependency relationship between annotation categories; and the annotation content node represents an annotation category.

3. The label category fusion method of claim 2, wherein, The traversing the basic annotation category text and the incremental annotation category text, taking the node connection mode of the full-amount annotation category text as a reference, fusing the annotation content nodes of the basic annotation category text and the incremental annotation category text, and forming the annotation content nodes and connection mode of the fused annotation category text comprises: checking each annotation content node in the basic annotation category text and the incremental annotation category text, taking the full-amount annotation category text as a reference, moving upwards or downwards in the full-amount annotation category text until the first stage start node and stage end node of the full-amount annotation category text are encountered, and adding the nodes found in the full-amount annotation category text and lines corresponding to the nodes to the fused annotation category text.

4. The method of claim 3, wherein, The step of deleting redundant stages in the annotation framework of the full-amount annotation category text according to the fusion annotation category text comprises: The step of deleting redundant stages in the annotation framework of the full-amount annotation category text according to the fusion annotation category text comprises:

5. The method of claim 4, wherein, The step of deleting redundant stages in the annotation framework of the full-amount annotation category text according to the fusion annotation category text comprises: The step of deleting redundant stages in the annotation framework of the full-amount annotation category text according to the fusion annotation category text comprises:

6. The method of claim 2, wherein, The method further comprises: traversing the full-amount annotation category text from the task starting node to form the annotation framework of the full-amount annotation category text by using the DSF algorithm.

7. The method of claim 1, wherein, The method further comprises: If the annotation category needs to be modified, the old annotation category in the basic annotation category text is ignored when forming the annotation content nodes and connection modes of the fusion annotation category text.

8. The method of claim 1, wherein, The node is set as (box / property_id, node_id, meta), wherein the box / property_id is an attribute label, the node_id is a node label, and the meta is an annotation content format defined by the node.

9. An annotated class fusion system, comprising: The system comprises: The processing module is configured to obtain a basic annotation category text and an incremental annotation category text, and maintain a full-amount annotation category text. The basic annotation category text, the incremental annotation category text and the full-amount annotation category text are provided with the same node type, structure and annotation content expression form, the node type comprises an annotation content node and a form node, the full-amount annotation category text comprises annotation content nodes of all supported annotation types in a business range and all form nodes, the node type and structure in the annotation category text are specified, that is, the nodes in the annotation category text are divided into fixed types, and the nodes of each type are provided with a corresponding structure, the form node type comprises a task starting node and a task ending node, a stage starting node and a stage ending node, and a clustering node, wherein the task starting node and the task ending node have no annotation meaning, but can specify the annotation category text structure as a tree without losing expression ability, and the forest is removed. a first fusion module, configured to traverse the basic annotation category text and the incremental annotation category text, and fuse annotation content nodes of the basic annotation category text and the incremental annotation category text in a node connection mode of the full-amount annotation category text to form annotation content nodes and connection modes of the fusion annotation category text, wherein the connection modes of the form nodes of the full-amount annotation category text are recorded in an annotation framework of the full-amount annotation category text; a second fusion module, configured to delete redundant stages in the annotation framework of the full-amount annotation category text according to the fusion annotation category text, form a new annotation framework, and combine the fusion annotation category text and the new annotation framework to form a fusion annotation category text.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the annotation category fusion method in any one of claims 1 to 8.

Images