Calculation amount drawing-oriented large model training data labeling method, device and equipment

Abstract

The invention provides a calculation amount drawing-oriented large model training data labeling method, device and equipment, which are used for labeling key information of a construction engineering cost calculation amount drawing. The method comprises the following steps: displaying a to-be-labeled calculation amount drawing, determining a to-be-identified area, and identifying to obtain text content and position information of a text block; displaying a label list according to the keyword field; after a user selects the tag, activating the interaction state of the to-be-identified area, so that the text block is presented as an operable hot area; after the user selects the hot area, generating a key value pair by using the selected label and the hot area text; and collecting the key value pairs of the to-be-identified areas to generate answer content, and filling the answer content into an answer part of a preset question-answer template to obtain a labeling result. By adopting the method, the field and text matching can be quickly completed, the structured annotation data is generated, the annotation efficiency and consistency are improved, and the method can be conveniently and directly used for large model training.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, and equipment for labeling training data for large models of quantity survey drawings. Background Technology

[0002] In the process of quantity surveying for construction projects, it is often necessary to identify and extract structured information such as components, dimensions, material symbols, and text descriptions from quantity survey drawings to support tasks such as quantity calculation, bill of quantities preparation, and cost analysis. Quantity survey drawings typically contain graphic sets of components, dimension annotations, material symbols, and extensive text descriptions, characterized by high information density, mixed text and graphics, diverse expression formats, and the possibility that the same keyword may correspond to multiple texts or various annotation methods. With the development of deep learning technology, multimodal large-scale models can simultaneously receive image pixels and text sequences from quantity survey drawings as input and output structured key information end-to-end, thereby improving the automation level of drawing understanding and key information extraction. To train such multimodal large-scale models, it is usually necessary to construct a dataset containing images, text, and structured labels, and the construction of the dataset depends on the accurate annotation of the original drawing data.

[0003] In existing technologies, image and text data annotation primarily relies on manual operations, with common tools including Labelme and VGG Image Annotator (VIA). For example, when using Labelme, users typically need to import the image into their local software, outline each individual text target one by one, manually fill in the text content in a pop-up window, and further manually add corresponding keyword fields. Then, a JSON file is exported. The exported annotation results still require post-processing to construct a key-value pair dataset suitable for model training. Similarly, when using VIA, users generally need to load the image in the browser and predefine keyword fields. After outlining the target, a keyword table pops up, and then users manually fill in the corresponding keyword text content and export a JSON file. Likewise, the exported results usually require post-processing to meet the training data format requirements.

[0004] To improve efficiency, some tools have incorporated automatic text recognition capabilities. For example, Baidu PaddleLabel can automatically detect and recognize all text in an image and export it as JSON. However, such solutions typically focus on the detection, recognition, and output of text, lacking keyword fields and key-value pair matching processes for key information extraction tasks. This makes it difficult to directly generate the structured labels required for multimodal model training, and users still need to manually filter, classify, and extract key-value pairs from the recognition results.

[0005] The inventors' research on existing annotation schemes revealed the following common problems: First, high labor costs, requiring manual selection of image regions and input of corresponding text, resulting in low overall efficiency; second, traditional annotation tools do not support or integrate OCR functionality, necessitating human visual recognition and manual input of text within images, which is error-prone and time-consuming; third, while schemes with OCR capabilities can output all text, they lack the key-value pair matching and template-based organization processes required for extracting key information, still necessitating additional manual extraction and post-processing. In summary, existing annotation tools struggle to achieve automatic processing and structured organization of text information, and the generated data often cannot be directly used for training multimodal large-scale models, requiring the additional construction of training data templates and post-processing.

[0006] Therefore, when annotating quantity survey drawings, how to reduce the workload of manual text input, improve the matching efficiency between keyword fields and text content, and enable the annotation results to be output in a structured form that meets the training requirements of multimodal large models has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0007] The purpose of this invention is to provide a method, apparatus, and device for annotating large model training data for quantity survey drawings, in order to solve the aforementioned technical problems in the prior art.

[0008] On the one hand, in order to achieve the above objectives, the present invention provides a method for labeling large model training data for quantity calculation drawings.

[0009] The large-scale model training data annotation for quantity survey drawings includes: displaying the quantity survey drawing to be annotated, wherein the quantity survey drawing to be annotated includes several keyword fields; determining the regions to be identified in the quantity survey drawing to be annotated; identifying the regions to be identified, obtaining the text content and position information of several text blocks; displaying a tag list according to the keyword fields, wherein the tag list includes tags corresponding to each keyword field; in response to the user's selection operation of the tags, recording the selected tags and activating the interactive state of the regions to be identified, wherein when the regions to be identified are in an interactive state, each text block of the regions to be identified is presented as an operable hotspot; in response to the user's selection operation of the hotspots, generating key-value pairs using the selected tags and the text content corresponding to the selected hotspots; generating answer content based on the key-value pairs corresponding to all the regions to be identified in the quantity survey drawing to be annotated; and filling the answer content into the answer part of a preset question-answer template to generate the annotation result of the quantity survey drawing to be annotated.

[0010] Furthermore, the step of activating the interactive state of the area to be identified includes: overlaying text content on the location marked by the location information in the area to be identified; and setting the location marked by the location information as the operable hot zone.

[0011] Further, the step of displaying the tag list based on the keyword field includes: displaying the tag and the corresponding content input box; the user's selection operation of the tag includes the selection operation of the content input box; the step of generating key-value pairs using the selected tag and the text content corresponding to the selected hotspot includes: filling the text content corresponding to the selected hotspot into the selected content input box, and generating the key-value pairs using the tag corresponding to the selected content input box and the text content in the selected content input box.

[0012] Furthermore, the step of overlaying the text content at the location indicated by the location information in the area to be identified includes: overlaying the text content as a semi-transparent highlighted area at the location indicated by the location information, so that the text content is displayed in a floating layer above the area to be identified.

[0013] Further, the step of determining the area to be identified in the quantity calculation drawing to be labeled includes: in response to the user's operation of the area delineation tool, drawing a closed outline on the quantity calculation drawing to be labeled to determine the area to be identified; the method also includes: in response to the user's operation of the area delineation tool, displaying a tag list according to the keyword field.

[0014] Further, the step of identifying the area to be identified and obtaining the text content and location information of several text blocks includes: after outlining a closed contour on the quantity calculation drawing to be labeled, cropping the closed contour and sending it to a text detection and recognition tool, wherein the text detection and recognition tool is used to detect the text in the closed contour, obtain the text block and the location information of the text block, crop the text block according to the location information and perform recognition to obtain the text content of the text block; and receiving the text content and location information of the text block returned by the text detection and recognition tool.

[0015] Furthermore, before displaying the quantity calculation drawing to be annotated, the method further includes: acquiring a quantity calculation drawing dataset, wherein the quantity calculation drawing dataset includes multiple quantity calculation drawings, each of which includes the same keyword field; displaying a keyword annotation page corresponding to the quantity calculation drawing to be annotated, wherein the keyword annotation page includes an attribute name input box and an attribute type input box; in response to the content entered by the user in the attribute name input box and the attribute type input box, generating an attribute list including the attribute name and attribute type, wherein the attribute name in the attribute list is the field name of the keyword field; wherein the step of displaying a tag list according to the keyword field includes: displaying the tag list according to the attribute list.

[0016] On the other hand, in order to achieve the above objectives, the present invention provides a large model training data annotation device for quantity calculation drawings.

[0017] The large-scale model training data annotation device for quantity survey drawings includes: a first display module for displaying the quantity survey drawing to be annotated, wherein the quantity survey drawing includes several keyword fields; a first determination module for determining the region to be identified in the quantity survey drawing; an identification module for identifying the region to be identified and obtaining the text content and position information of several text blocks; a second display module for displaying a tag list according to the keyword fields, wherein the tag list includes tags corresponding to each keyword field; and a first response module for responding to a user's selection operation of the tags, recording the selected tags, and... The interactive state of the area to be identified is activated, wherein when the area to be identified is in an interactive state, each text block of the area to be identified is presented as an operable hotspot; a second response module is used to respond to the user's selection operation of the hotspot and generate key-value pairs using the selected label and the text content corresponding to the selected hotspot; a first generation module is used to generate answer content based on the key-value pairs corresponding to all the areas to be identified in the quantity calculation drawing to be labeled; and a second generation module is used to fill the answer content into the answer part of the preset question-answer template to generate the labeling result of the quantity calculation drawing to be labeled.

[0018] On the other hand, to achieve the above objectives, the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above method.

[0019] On the other hand, to achieve the above objectives, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above method.

[0020] This invention provides a method, apparatus, and device for annotating large-scale model training data for quantity survey drawings. The method displays the quantity survey drawing to be annotated and identifies the regions to be identified. These regions are then identified to obtain text content and their location information. A tag list is displayed based on keyword fields in the drawing. After the user selects a tag, the interactive state of the regions to be identified is activated, making the text blocks operable hotspots. Corresponding key-value pairs are generated after the user selects a hotspot. The key-value pairs from each region to be identified are then aggregated to generate answer content, which is then filled into a preset question-answer template to generate the annotation result. This invention combines tag selection with hotspot selection to match keyword fields with text content and outputs the annotation result through a question-answer template. This reduces the workload of manual text entry and data organization, improves annotation efficiency and consistency, and makes the output results easier to use directly for large-scale model training data construction. Attached Figure Description

[0021] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0022] Figure 1 This is a flowchart of a method for labeling training data for large models of quantity calculation drawings provided in Embodiment 1 of the present invention; Figure 2 A schematic diagram of a dataset management interface provided in an embodiment of the present invention; Figure 3 A schematic diagram of a keyword annotation interface provided in an embodiment of the present invention; Figure 4 A schematic diagram of a text annotation interface provided in an embodiment of the present invention; Figure 5 Another schematic diagram of the text annotation interface provided in the embodiments of the present invention; Figure 6 This is a schematic diagram of the QA generation interface provided in an embodiment of the present invention; Figure 7 This is a block diagram of a large model training data annotation device for quantity calculation drawings provided in Embodiment 2 of the present invention; Figure 8 This is a hardware structure diagram of a computer device provided in Embodiment 3 of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0024] Example 1 This invention provides a method for annotating large-scale model training data for quantity survey drawings. This method interactively matches text content in the area to be identified with keyword fields in the quantity survey drawings during the annotation process. This reduces the workload of manual text input while improving the matching efficiency between keyword fields and text content, and generates annotation results that can be used for large-scale model training. Specifically, Figure 1 The flowchart of the large model training data annotation method for quantity calculation drawings provided in Embodiment 1 of the present invention is as follows: Figure 1 As shown, the method provided in this embodiment includes the following steps S101 to S108.

[0025] Step S101: Display the quantity calculation drawing to be annotated.

[0026] The quantity calculation drawings are design drawings used for quantity calculation in building construction projects. They may include component assembly graphics, dimensions, material symbols, and text descriptions. These contents can be abstracted into several fields. Therefore, some or all of these fields are defined as keyword fields to represent the key information that needs to be extracted from the quantity calculation drawings to be annotated, such as component name, specifications, material type, dimensional parameters, and quantity information. Optionally, the keyword fields of the quantity calculation drawings to be annotated can be set manually, or a database can be set up to correspond to the keyword fields and the quantity calculation drawing types. For a specific quantity calculation drawing to be annotated, the corresponding keyword fields are determined from the database based on its quantity calculation drawing type.

[0027] Step S102: Determine the area to be identified in the quantity calculation drawing to be annotated.

[0028] In this embodiment, the area to be identified is used to define a local area of ​​the drawing that needs to be detected and recognized by text, that is, the unit for text detection and recognition. Optionally, the area to be identified may correspond to a functional area, a component description area, a densely labeled area, or a frame / table area in the drawing, so that the text in the area can be identified and used for annotation later. Optionally, the area to be identified can be manually labeled or automatically detected by machine; there is no limitation on this.

[0029] Step S103: Identify the region to be identified and obtain the text content and location information of several text blocks.

[0030] In this embodiment, recognition processing is performed on the region to be recognized to obtain the text content and location information corresponding to multiple text blocks in the region. Each text block is a relatively independent text region within the region to be recognized. The location information is used to characterize the position of the text block within the region, such as using bounding box coordinates, corner coordinates, or polygon outline coordinates, thereby supporting subsequent location-based visualization and interactive selection.

[0031] Step S104: Display the tag list based on the keyword field.

[0032] In this embodiment, the tag list serves as the interface between the keyword field and the annotation interaction. The tag list includes tags corresponding to each keyword field, ensuring a one-to-one correspondence between each tag and a keyword field, making it easier for users to clearly identify the target field to be annotated.

[0033] Step S105: In response to the user's selection of a label, record the selected label and activate the interactive state of the area to be identified.

[0034] In this embodiment, when a user selects a tag, the currently selected tag is recorded to indicate the "key" of the key-value pair to be generated. Furthermore, the interactive state of the area to be identified is activated: when the area is interactive, each text block within the area is presented as an operable hotspot, allowing the user to quickly specify the source text of the "value" through selection of the hotspot.

[0035] Step S106: In response to the user's selection of a hotspot, generate key-value pairs using the selected label and the text content corresponding to the selected hotspot.

[0036] In this embodiment, when a user selects a hotspot, the text content corresponding to that hotspot is obtained, and this text content is combined with the selected tag recorded in step S105 to generate a key-value pair. The key of the key-value pair is the name of the keyword field corresponding to the selected tag, and the value of the key-value pair is the text content corresponding to the selected hotspot, thereby establishing a correspondence between the keyword field and the corresponding text content in the drawing.

[0037] Step S107: Generate the answer content based on the key-value pairs corresponding to all areas to be identified in the quantity calculation drawing to be annotated.

[0038] In this embodiment, for the same quantity calculation drawing to be labeled, there may be one or more regions to be identified. The key-value pairs generated for each region to be identified are aggregated to generate corresponding answer content. The answer content is used to summarize the labeling results of the drawing in the keyword field dimension, so as to uniformly write it into the subsequent preset question-answer template. The preset question-answer template is used to organize the labeled data into a data format oriented towards large model training. The preset question-answer template includes a question part and an answer part; the question part can be preset content.

[0039] Step S108: Fill in the answer content into the answer section of the preset question-answer template to generate the annotation result of the quantity calculation drawing to be annotated.

[0040] In this embodiment, the answer content generated in step S107 is filled into the answer part of the preset question-answer template to obtain the complete question part and answer part, and the annotation result of the quantity calculation drawing to be annotated is output, so that the annotation result can be used for subsequent large model training data construction and management.

[0041] In the large-scale model training data annotation method for quantity survey drawings provided in this embodiment, the quantity survey drawing to be annotated is displayed, and the regions to be identified are determined. These regions are then identified to obtain text content and their location information. A tag list is displayed based on the keyword fields of the quantity survey drawing. After the user selects a tag, the interactive state of the regions to be identified is activated, making the text blocks operable hotspots. Corresponding key-value pairs are generated after the user selects a hotspot. The key-value pairs of each region to be identified are further aggregated to generate answer content, which is then filled into a preset question-answer template to generate the annotation result. Using the large-scale model training data annotation method for quantity survey drawings provided in this embodiment, the matching of keyword fields and text content can be completed by combining tag selection and hotspot selection. The annotation result is output through a question-answer template, thereby reducing the workload of manual text entry and manual organization of training data, improving annotation efficiency and consistency, and making the output results easier to directly use for large-scale model training data construction.

[0042] Optionally, in one embodiment, before displaying the quantity calculation drawing to be annotated, the annotation method further includes: obtaining a quantity calculation drawing dataset, wherein the quantity calculation drawing dataset includes multiple quantity calculation drawings, each of which includes the same keyword field; displaying a keyword annotation page corresponding to the quantity calculation drawing to be annotated, wherein the keyword annotation page includes an attribute name input box and an attribute type input box; in response to the content entered by the user in the attribute name input box and the attribute type input box, generating an attribute list including attribute names and attribute types, wherein the attribute names in the attribute list are the field names of the keyword field; wherein the step of displaying a tag list according to the keyword field includes: displaying a tag list according to the attribute list.

[0043] Specifically, Figure 2 This is a schematic diagram of the dataset management interface provided in an embodiment of the present invention, such as... Figure 2 As shown, datasets and versions can be created in the dataset management interface, and multiple images or image compressed packages can be imported to obtain a quantity calculation drawing dataset. The dataset management interface displays all imported quantity calculation drawings, as well as statistics on labeled and unlabeled quantity calculation drawings. Each quantity calculation drawing package in the dataset shares the same keyword field, thus allowing for unified labeling of quantity calculation drawings with the same attribute. This enables the labeling of multiple quantity calculation drawings of the same type, generating training data of the same type.

[0044] Figure 3 This is a schematic diagram of the keyword annotation interface provided in an embodiment of the present invention, such as... Figure 3 As shown, after importing the quantity calculation drawing dataset, you can enter the keyword annotation page of this dataset to complete the customization of the keyword field. The keyword annotation page displays an attribute name input box and an attribute type input box. The user enters the attribute name in the attribute name input box, such as: thermal insulation type, corrosion protection type, design pressure, thermal insulation thickness and diameter, etc., and enters the attribute type in the attribute type input box, such as the attribute type including text.

[0045] Based on the user's input in the attribute name and attribute type input boxes on this interface, an attribute list including attribute name and attribute type is generated. The attribute list allows for unified management of keyword field configurations, enabling multiple quantity calculation drawings within the same dataset to share a consistent set of fields. The attribute type can be used to characterize the attribute category of a field, thus supporting subsequent organization of a label list by field. Each label in the label list corresponds to each attribute name in the attribute list.

[0046] In the large model training data annotation method for quantity calculation drawings provided in this embodiment, the label list can be automatically displayed based on a unified set of fields by configuring keyword fields and generating attribute lists at the dataset level. This reduces the workload of repeatedly configuring fields in multi-drawing scenarios, improves the consistency and standardization of batch annotation, and promotes the structural consistency of annotation results under the same question-answer template, thereby making it easier to uniformly construct and manage large model training data.

[0047] Optionally, in one embodiment, the step of determining the area to be identified in the quantity calculation drawing to be annotated includes: in response to the user's operation of the area delineation tool, drawing a closed outline on the quantity calculation drawing to be annotated to determine the area to be identified; the annotation method further includes: in response to the user's operation of the area delineation tool, displaying a tag list based on the keyword field.

[0048] Specifically, Figure 4This is a schematic diagram of a text annotation interface provided in an embodiment of the present invention. Figure 5 Another schematic diagram of the text annotation interface provided in the embodiments of the present invention, such as... Figure 4 and Figure 5 As shown, the text annotation page provides an intuitive interactive interface when users annotate a single quantity calculation drawing. A region delineation tool is provided to allow users to define the local area to be identified on the displayed quantity calculation drawing. Users can use the rectangle tool to draw one or more closed outlines on the image of the quantity calculation drawing to be annotated. Each outline represents a logical text group, corresponding to a region to be identified, such as... Figure 4 The two regions to be identified shown are region 1 and region 2. A list of labels is displayed synchronously when the outlining operation occurs, such as... Figure 4 The right side displays the tag lists corresponding to Area 1 and Area 2 respectively. The areas and tag lists correspond one-to-one. They can be displayed on the same page at the same time or sequentially on multiple pages, so that users can directly enter the tag selection and hotspot selection process after completing the area selection.

[0049] In the large model training data annotation method for quantity calculation drawings provided in this embodiment, the area to be identified is quickly determined by delineating the closed contour, and a label list is displayed when the delineation operation is triggered, thereby shortening the operation link between area determination and field matching, and improving the continuity and overall efficiency of the annotation process.

[0050] Optionally, in one embodiment, the step of identifying the area to be identified and obtaining the text content and location information of several text blocks includes: after outlining a closed contour on the quantity calculation drawing to be annotated, cutting the closed contour and sending it to a text detection and recognition tool, wherein the text detection and recognition tool is used to detect the text in the closed contour, obtain the text blocks and the location information of the text blocks, cut the text blocks according to the location information and perform recognition to obtain the text content of the text blocks; and receive the text content and location information of the text blocks returned by the text detection and recognition tool.

[0051] Specifically, continuing with example 4, after the user completes and confirms the outline, the front end automatically crops the image area corresponding to the outline and sends it to the back end text detection and recognition tool. Through the overall process of the text detection and recognition tool, the detection of the area to be identified in the quantity survey drawing is achieved. The text detection and recognition tool includes a document detection model and a text recognition model. Both models are trained on a large number of quantity survey drawings, ensuring high accuracy in detecting and recognizing the area to be identified in the quantity survey drawings. After the cropped area enters the text detection and recognition tool, it first passes through the text detection model to obtain the position coordinates of multiple lines of text blocks in the cropped area. The position information of the cropped area and text blocks is then sent to the text recognition model. The text recognition model further crops and corrects the text blocks based on their position information before performing batch recognition, outputting the text content and confidence score of each text block. The position information of the text blocks and their corresponding text content are packaged and sent back to the front end. The front end then obtains the position information of all text blocks within the area to be identified and their recognized text content. By cropping the closed contour area and performing detection and recognition only on the cropped result, the recognition task can be limited to the area to be recognized. Furthermore, by cropping the text block based on the detected position information and then performing recognition, the text content and position can be output together to support subsequent hotspot overlay and point selection interaction.

[0052] In the large model training data annotation method for quantity calculation drawings provided in this embodiment, the processing method of combining region clipping with text detection and recognition tools to output text content and location information enables text blocks in the area to be identified to be extracted in a structured manner and used for hot zone interaction, reducing interference from irrelevant areas and improving the stability and efficiency of recognition and annotation.

[0053] Optionally, in one embodiment, the step of activating the interactive state of the region to be identified includes: overlaying text content on the location information marked in the region to be identified; and setting the location information marked as an operable hotspot.

[0054] Specifically, such as Figure 5 As shown, when the interactive state of the area to be recognized is activated, the text content is directly overlaid and displayed at the location marked in the area to be recognized. Specifically, a display method different from the original display method of the area to be recognized can be used, such as using different colors, so that the recognized text content of each text block is directly visible in its corresponding position, making it easy for users to quickly locate and verify the recognized text content. At the same time, the location marked by the location information is set as a hot zone. Through the setting of the hot zone, users can directly click on the text block in the area to be recognized.

[0055] The large model training data annotation method for quantity calculation drawings provided in this embodiment overlays text in the original position and sets hot zones to achieve synchronous operations of positioning, verification and selection, thereby realizing interactive annotation of what you see is what you select. Users can input content by selecting while verifying, reducing the user's operation cost when matching tags, reducing wrong selection and wrong filling, and further improving the accuracy of field matching and the efficiency of annotation operation.

[0056] Optionally, in one embodiment, the step of displaying a tag list based on a keyword field includes: displaying tags and corresponding tag content input boxes; the user's selection operation on the tags includes a selection operation on the content input boxes; and the step of generating key-value pairs using the selected tags and the text content corresponding to the selected hotspots includes: filling the text content corresponding to the selected hotspots into the selected content input boxes, and generating key-value pairs using the tags corresponding to the selected content input boxes and the text content in the selected content input boxes.

[0057] Specifically, as the user outlines the target, the interface simultaneously updates the tag list. For example... Figure 5 As shown, based on the keyword fields corresponding to the current dataset, the right panel dynamically generates corresponding label items. Each label displays a blank input box as its corresponding content input box, used to hold the fill value for the field corresponding to that label. Users select the current field by clicking the input box, and then automatically fill in the corresponding text content by clicking the hotspot. Optionally, a "two-click" interaction protocol is used to achieve rapid construction of key-value pairs. First click: In the initial state, all content input boxes await user-triggered filling operations. User selection of labels includes selection of content input boxes. When a user clicks a content input box, that box enters a "pending fill" state, recording the currently selected keyword field. Simultaneously, the interactive state of all pre-labeled text blocks in the left image is activated, and all pre-labeled text blocks in the left image enter a selectable highlighted state. Each text block area is presented as a clickable hotspot, prompting the user to perform association operations. Second click: The user clicks the hotspot of the text block matching the current keyword in the left image, automatically filling the corresponding text content of that text block into the right text input box, completing the labeling of a key-value pair. After each fill, the annotation progress is updated in real time, and "undo" or "clear" operations are provided, allowing users to reselect or manually modify the text content. Users can continue clicking or using the tab key to switch to the next keyword input box, repeating the above two-click process until all keywords in the area to be identified are annotated. By repeating the above "two-click" interaction for each area to be identified, the annotation of the current quantity calculation drawing can be completed, resulting in multiple sets of key-value pairs.

[0058] In the large model training data annotation method for quantity calculation drawings provided in this embodiment, labels are bound to content input boxes, and content is automatically filled and key-value pairs are generated through hot zone selection, thereby reducing copy-paste and manual input operations, reducing the probability of human input errors, and improving the consistency of field-text matching and annotation efficiency.

[0059] Optionally, in one embodiment, the step of overlaying text content at the location marked by the location information in the area to be identified includes: overlaying the text content as a semi-transparent highlighted area at the location marked by the location information, so that the text content is displayed in a floating layer above the area to be identified.

[0060] Specifically, semi-transparent highlighted areas are used to emphasize the range of text blocks, while overlay displays are used to show the identified text content without obscuring the details of the original drawing. In the large model training data annotation method for quantity calculation drawings provided in this embodiment, the semi-transparent highlighting and overlay display methods balance readability and drawing visibility, reduce the obscuring of the original drawing content by the overlay display, improve the user-friendliness of hotspot recognition and point selection operations, and thus further improve annotation efficiency and operational accuracy.

[0061] Optionally, in one embodiment, after obtaining the key-value pairs for each region to be identified, the question-answer pair annotation page of this dataset is accessed. Figure 6 This is a schematic diagram of the QA generation interface provided in an embodiment of the present invention, as shown below. Figure 6 As shown, using the Jinja template engine, structured data can be injected into a preset Prompt-Response template with a single click, such as... Figure 6 The QA generation configuration section is shown, and the final result is rendered in real time, as shown. Figure 6 The preview section showing the QA generation results is shown.

[0062] In summary, the large-scale model training data annotation method for quantity survey drawings provided in this embodiment can improve the annotation efficiency of key information extraction datasets. It replaces mouse dragging or manual typing with two-click copy-paste, reducing the annotation time for a single image from the traditional 3-5 minutes to 20-40 seconds, an efficiency improvement of 5-8 times. By providing unified character-level results through pre-annotation, manual modification or supplementation is possible, eliminating differences in spelling, formatting, and abbreviation habits among different annotators. The entire process of text detection, recognition, key information extraction, and field backfilling can be completed within the same system, and the output can be used for large-scale model training without secondary cleaning or annotation, saving manpower and time and significantly shortening the project cycle. In this application, after the user outlines any closed contour, the text detection and recognition tool is only called on that local pixel, returning the coordinates and content of multiple lines of text blocks at once as the sole source for subsequent key-value matching, avoiding noise from full-image recognition. The front end renders the pre-annotated text blocks as invisible hotspots, each hotspot carrying a string, i.e., the corresponding text content. The first click on the blank input box on the right selects the pre-filled area, and a hotspot appears. The second click on the hotspot text block allows for copying and pasting. Two clicks complete one key-value pair. The Jinja2 question-answering template engine is embedded, and within the same annotation session, all annotated key-value pairs are injected into the Jinja2 template in real time, instantly generating the Prompt-Response instruction pairs required for training multimodal large models, so that training samples are ready as soon as annotation is completed.

[0063] Example 2 Corresponding to Embodiment 1 above, Embodiment 2 of the present invention provides a large model training data annotation device for quantity calculation drawings. The technical features and corresponding technical effects can be referred to Embodiment 1 above, and will not be repeated in this embodiment. Figure 7 This is a block diagram of the large model training data annotation device for quantity calculation drawings provided in Embodiment 2 of the present invention, as follows: Figure 7 As shown, the device includes a first display module 201, a first determination module 202, an identification module 203, a second display module 204, a first response module 205, a second response module 206, a first generation module 207, and a second generation module 208.

[0064] The system comprises the following components: a first display module 201 for displaying a quantity calculation drawing to be annotated, wherein the quantity calculation drawing includes several keyword fields; a first determination module 202 for determining the area to be identified in the quantity calculation drawing; a recognition module 203 for recognizing the area to be identified and obtaining the text content and location information of several text blocks; a second display module 204 for displaying a tag list based on the keyword fields, wherein the tag list includes tags corresponding to each keyword field; and a first response module 205 for responding to a user's selection of a tag, recording the selected tag, and activating the area to be identified. The interactive state of the area, wherein when the area to be identified is in an interactive state, each text block of the area to be identified is presented as an operable hotspot; the second response module 206 is used to respond to the user's selection operation of the hotspot and generate key-value pairs using the selected label and the text content corresponding to the selected hotspot; the first generation module 207 is used to generate answer content based on the key-value pairs corresponding to all the areas to be identified in the quantity calculation drawing to be labeled; and the second generation module 208 is used to fill the answer content into the answer part of the preset question-answer template to generate the labeling result of the quantity calculation drawing to be labeled.

[0065] Optionally, in one embodiment, the first response module includes: a first display unit, configured to overlay text content onto the location indicated by the location information in the area to be identified; and a first control unit, configured to set the location indicated by the location information as the operable hot zone.

[0066] Optionally, in one embodiment, the second display module is further configured to display the label and a content input box corresponding to the label; the user's selection operation on the label includes a selection operation on the content input box; the second response module includes: a filling unit, configured to fill the text content corresponding to the selected hot zone into the selected content input box; and a generating unit, configured to generate the key-value pair using the label corresponding to the selected content input box and the text content in the selected content input box.

[0067] Optionally, in one embodiment, the first display unit is further configured to overlay the text content as a semi-transparent highlighted area at the position marked by the location information, so that the text content is displayed in a floating layer above the area to be identified.

[0068] Optionally, in one embodiment, the first determining module is further configured to, in response to the user's operation of the area delineation tool, delineate a closed outline on the quantity calculation drawing to be labeled, and determine the area to be identified; the device further includes: a third response module, configured to, in response to the user's operation of the area delineation tool, display a tag list according to the keyword field.

[0069] Optionally, in one embodiment, the recognition module includes: a processing unit, configured to, after outlining a closed contour on the quantity calculation drawing to be annotated, cut the closed contour and send it to a text detection and recognition tool, wherein the text detection and recognition tool is configured to detect text in the closed contour, obtain text blocks and position information of the text blocks, cut the text blocks according to the position information and perform recognition to obtain the text content of the text blocks; and a receiving unit, configured to receive the text content and position information of the text blocks returned by the text detection and recognition tool.

[0070] Optionally, in one embodiment, the apparatus further includes: an acquisition module, configured to acquire a quantity calculation drawing dataset before the first display module displays the quantity calculation drawing to be labeled, wherein the quantity calculation drawing dataset includes multiple quantity calculation drawings, and each quantity calculation drawing includes the same keyword field; a third display module, configured to display a keyword labeling page corresponding to the quantity calculation drawing to be labeled, wherein the keyword labeling page includes an attribute name input box and an attribute type input box; a fourth response module, configured to generate an attribute list including the attribute name and attribute type in response to the content entered by the user in the attribute name input box and the attribute type input box, wherein the attribute name in the attribute list is the field name of the keyword field; wherein the second display module is further configured to display the tag list according to the attribute list.

[0071] Example 3 This embodiment also provides a computer device, such as a smartphone, tablet computer, laptop computer, desktop computer, rack server, blade server, tower server, or cabinet server (including a standalone server or a server cluster composed of multiple servers), etc., capable of executing programs. Figure 8 As shown, the computer device 01 in this embodiment includes, but is not limited to, a memory 012 and a processor 011 that can be interconnected via a system bus, such as... Figure 8 As shown. It should be noted that, Figure 8 Only a computer device 01 with component memory 012 and processor 011 is shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0072] In this embodiment, the memory 012 (i.e., the readable storage medium) includes flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 012 may be an internal storage unit of the computer device 01, such as the hard disk or memory of the computer device 01. In other embodiments, the memory 012 may also be an external storage device of the computer device 01, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 01. Of course, the memory 012 may include both the internal storage unit and its external storage device of the computer device 01. In this embodiment, the memory 012 is typically used to store the operating system and various reference software installed on the computer device 01, such as the program code of the large model training data annotation device for quantity calculation drawings in Embodiment 2. In addition, memory 012 can also be used to temporarily store various types of data that have been output or will be output.

[0073] In some embodiments, processor 011 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. Processor 011 is typically used to control the overall operation of computer device 01. In this embodiment, processor 011 is used to run program code stored in memory 012 or process data, such as a method for annotating large model training data for quantity surveying drawings.

[0074] Example 4 This embodiment also provides a computer-readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, server, App reference store, etc., which stores a computer program. When the program is executed by a processor, it implements the corresponding function. The computer-readable storage medium of this embodiment is used to store a large model training data annotation device for quantity survey drawings. When executed by a processor, it implements the large model training data annotation method for quantity survey drawings in Embodiment 1.

[0075] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0076] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0077] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0078] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for labeling training data for large models of quantity survey drawings, characterized in that, include: Display the quantity calculation drawing to be annotated, wherein the quantity calculation drawing to be annotated includes several keyword fields; Identify the areas to be identified in the quantity calculation drawings to be annotated; The region to be identified is identified to obtain the text content and location information of several text blocks; Display a tag list based on the keyword field, wherein the tag list includes tags corresponding to each keyword field; In response to the user's selection of the tag, the selected tag is recorded, and the interactive state of the area to be identified is activated. When the area to be identified is in an interactive state, each text block of the area to be identified is presented as an operable hotspot. In response to the user's selection of the hot zone, a key-value pair is generated using the selected tag and the text content corresponding to the selected hot zone; The answer content is generated based on the key-value pairs corresponding to all the areas to be identified in the quantity calculation drawing to be labeled; and Fill the answer content into the answer section of the preset question-answer template to generate the annotation result of the quantity calculation drawing to be annotated.

2. The method for labeling large model training data for quantity survey drawings according to claim 1, characterized in that, The steps to activate the interactive state of the region to be identified include: The text content is overlaid and displayed at the location indicated by the location information in the area to be identified; The location identified by the location information is set as the operable hot zone.

3. The method for annotating large model training data for quantity survey drawings according to claim 2, characterized in that, The step of displaying a tag list based on the keyword field includes: displaying the tag and a content input box corresponding to the tag; The user's selection of the label includes the selection of the content input box; The step of generating key-value pairs using the selected label and the text content corresponding to the selected hotspot includes: filling the text content corresponding to the selected hotspot into the selected content input box, and generating the key-value pairs using the label corresponding to the selected content input box and the text content in the selected content input box.

4. The method for labeling large model training data for quantity survey drawings according to claim 2, characterized in that, The step of overlaying text content onto the location indicated by the location information in the area to be identified includes: The text content is overlaid as a semi-transparent highlighted area at the location marked by the location information, so that the text content is displayed in a floating layer above the area to be identified.

5. The method for annotating large model training data for quantity survey drawings according to claim 1, characterized in that, The steps for determining the region to be identified in the quantity calculation drawing to be annotated include: in response to the user's operation of the region delineation tool, drawing a closed outline on the quantity calculation drawing to be annotated to determine the region to be identified; The method further includes: in response to the user's operation on the region delineation tool, displaying a list of tags based on the keyword field.

6. The method for labeling large model training data for quantity survey drawings according to claim 5, characterized in that, The steps of identifying the region to be identified and obtaining the text content and location information of several text blocks include: After outlining a closed contour on the quantity calculation drawing to be annotated, the closed contour is cropped and sent to a text detection and recognition tool. The text detection and recognition tool is used to detect the text in the closed contour, obtain the text block and the position information of the text block, crop the text block according to the position information and perform recognition to obtain the text content of the text block. Receive the text content and location information of the text block returned by the text detection and recognition tool.

7. The method for labeling large model training data for quantity survey drawings according to claim 1, characterized in that, Before displaying the quantity calculation drawing to be annotated, the method further includes: Obtain a quantity calculation drawing dataset, wherein the quantity calculation drawing dataset includes multiple quantity calculation drawings, and each quantity calculation drawing includes the same keyword field; Display the keyword annotation page corresponding to the quantity calculation drawing to be annotated, wherein the keyword annotation page includes an attribute name input box and an attribute type input box; In response to the content entered by the user in the attribute name input box and the attribute type input box, an attribute list including the attribute name and attribute type is generated, wherein the attribute name in the attribute list is the field name of the keyword field; The step of displaying the tag list based on the keyword field includes: displaying the tag list based on the attribute list.

8. A model training data annotation device for quantity survey drawings, characterized in that, include: The first display module is used to display the quantity calculation drawing to be annotated, wherein the quantity calculation drawing to be annotated includes several keyword fields; The first determining module is used to determine the area to be identified in the quantity calculation drawing to be labeled; The recognition module is used to recognize the region to be recognized and obtain the text content and location information of several text blocks; The second display module is used to display a tag list according to the keyword field, wherein the tag list includes tags corresponding to each keyword field; The first response module is used to respond to the user's selection operation of the tag, record the selected tag, and activate the interactive state of the area to be identified. When the area to be identified is in the interactive state, each text block of the area to be identified is presented as an operable hotspot. The second response module is used to respond to the user's selection operation of the hot zone by generating key-value pairs using the selected tag and the text content corresponding to the selected hot zone. The first generation module is used to generate answer content based on the key-value pairs corresponding to all the areas to be identified in the quantity calculation drawing to be annotated; and The second generation module is used to fill the answer content into the answer part of the preset question-answer template and generate the annotation result of the quantity calculation drawing to be annotated.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images