Search

CN-121999494-A - Print description and print coordinate data preparation and marking method, electronic device and storage medium

CN121999494ACN 121999494 ACN121999494 ACN 121999494ACN-121999494-A

Abstract

The invention discloses a printing description and printing coordinate data preparation and labeling method, electronic equipment and a storage medium, wherein a multi-class initial data set is constructed through templatized synthesis and real data acquisition, a SegGPT/SAM (sample preparation/sample preparation) is used for dividing and generating a pixel-level Mask and a normalized coordinate, and a multi-printing instance is separated and a hierarchical relationship is labeled; establishing a long-medium-short three-level description template and style system, performing LLM color rendering and consistency verification, performing standardized packaging after multi-dimensional data enhancement, and outputting training data by combining version management and quality evaluation. The method has the advantages of improving the efficiency by more than 10 times, improving the labeling accuracy by more than 95%, reducing the cost by 70%, supporting the expansion of multiple categories and providing a high-quality data base for the training of the vision-language model.

Inventors

  • LIN JIEXING
  • LIU PENG
  • LIN HANQUAN

Assignees

  • 厦门灵图科技有限公司

Dates

Publication Date
20260508
Application Date
20251230

Claims (8)

  1. 1. A print description and print coordinate data preparation and labeling method, comprising: step 1, constructing an initial data set containing a plurality of samples by combining templatization synthesis and real data acquisition, and establishing a metadata management system of 'major class-sub class-specific class', and labeling attribute labels for each class; Step 2, performing pixel-level segmentation on the printing areas of the samples in the initial dataset by using a segmentation model to generate Mask images, calculating the minimum circumscribed rectangle of the printing areas based on segmentation results to obtain boundary frame coordinates, and normalizing the boundary frame coordinates to a [0,1] interval; Step 3, a description template library and a style description system are established, wherein the description template library comprises three-level description templates of a long text template, a medium text template and a short text template, and the style description system comprises style labels of style classification, technique description and color description; step 4, carrying out data enhancement on the sample, and carrying out quality check and standardization and formatting of output data; And 5, performing systematic data management.
  2. 2. The print description and print coordinate data preparation and labeling method according to claim 1, wherein said step 1 is specifically: The method comprises the steps of 1.1, establishing a multi-product commodity template library, generating printing by utilizing a parameterization control mode through a programmed synthesis engine based on the commodity template library, and enriching and generating printing patterns, layout and background of the printing by adopting a random combination strategy and a diversity control module; 1.2, collecting real commodity images, and automatically screening qualified samples meeting preset standards through a quality detection algorithm; 1.3, establishing a three-level classification system of 'major class-sub class-specific class', and labeling attribute labels for each class for metadata management.
  3. 3. The print description and print coordinate data preparation and labeling method according to claim 1, wherein: The step 2 uses SegGPT or SAM segmentation models to segment the printed region to generate Mask images.
  4. 4. The print description and print coordinate data preparation and labeling method according to claim 1, wherein: in the step 3, the long text template contains information of features, styles, techniques, layouts and positions, the medium text template reserves key visual features and styles, and the short text template is used for core features.
  5. 5. The print description and print coordinate data preparation and labeling method according to claim 1, wherein said step 4 is specifically: the method comprises the steps of carrying out multi-dimensional data enhancement through geometric transformation, color enhancement and background processing, automatically detecting labeling abnormality through a cross-validation algorithm, carrying out consistency check, establishing a unified data standard format, and packaging all information into a training data packet.
  6. 6. The print description and print coordinate data preparation and labeling method according to claim 1, wherein said step 5 is specifically: The method comprises the steps of establishing a data version management mechanism, tracking iteration history of a data set, supporting version rollback, establishing a multi-index evaluation system comprising integrity, accuracy and diversity, generating a quality evaluation report, and providing training data output, an API interface and a visualization tool to facilitate training and deployment of a large language model.
  7. 7. An electronic device comprising a processor, a memory and an application program, said application program being stored in said memory and configured to perform the decal description and decal coordinate data preparation and labeling method according to any one of claims 1 to 6 by said processor.
  8. 8. A computer-readable storage medium, characterized in that the storage medium stores a computer program which, when executed in the computer, causes the computer to perform the decal description and decal coordinate data preparation and labeling method according to any one of claims 1 to 6.

Description

Print description and print coordinate data preparation and marking method, electronic device and storage medium Technical Field The invention belongs to the technical field of computer vision, and particularly relates to a printing description and printing coordinate data preparation and marking method, electronic equipment and a storage medium. Background In the field of intelligent design of clothing and daily necessities, a printing coordinate prediction and description generation technology based on a visual-language model (VLM) is highly dependent on high-quality, multi-dimensional and structured training data. However, current data preparation methods face four major core challenges: (1) Data scarcity and high cost The printing patterns of multiple commodities (such as T-shirts, bed sheets, throw pillows and the like) are complex and various, the traditional manual labeling mode is low in efficiency, the labeling cost of a single sample is high, the labeling consistency is difficult to ensure, and high-quality training data are difficult to obtain on a large scale. (2) Concept complexity and multi-granularity description deletion The VLM model needs to have both the overall recognition capability of "coarse granularity" and the local understanding capability of "fine granularity", however, the existing dataset lacks a systematic multi-granularity text description system, and it is difficult to simultaneously meet the long text requirement of the VLM for detailed information including style, technique, layout and the like in the training stage, and the short text requirement of concise and core information in the application stage. (3) Multiple printing scene processing and marking precision difficult problem A single image may contain multiple printed areas and the morphological differences of different merchandise classes are significant. The existing labeling method is difficult to uniformly and accurately manage the complex relationship between the pixel-level coordinates and the corresponding descriptions of the multiple printing examples, and confusion or inaccurate positioning among the multiple printing examples are easy to cause. (4) Insufficient data expansibility and uneven quality The traditional labeling flow lacks a systematic and automatic data generation and verification mechanism, and is difficult to quickly adapt to the expansion requirements of new products and new styles. In addition, subjective differences exist in manual labeling, which seriously affect the training effect and iteration speed of the VLM model. Disclosure of Invention The invention mainly aims to provide a printing description and printing coordinate data preparation and marking method, electronic equipment and a storage medium, and the technical scheme of combining templated synthesis and intelligent marking is used for systematically solving the problems of data scarcity, insufficient description granularity, difficult marking of multiple printing scenes, expansibility limitation and the like faced by the prior art, and providing a high-quality, diversified and standardized data basis for VLM training. In order to achieve the above object, one of the solutions of the present invention is: A printing description and printing coordinate data preparation and labeling method comprises the following steps: step 1, constructing an initial data set containing a plurality of samples by combining templatization synthesis and real data acquisition, and establishing a metadata management system of 'major class-sub class-specific class', and labeling attribute labels for each class; Step 2, performing pixel-level segmentation on the printing areas of the samples in the initial dataset by using a segmentation model to generate Mask images, calculating the minimum circumscribed rectangle of the printing areas based on segmentation results to obtain boundary frame coordinates, and normalizing the boundary frame coordinates to a [0,1] interval; Step 3, a description template library and a style description system are established, wherein the description template library comprises three-level description templates of a long text template, a medium text template and a short text template, and the style description system comprises style labels of style classification, technique description and color description; step 4, carrying out data enhancement on the sample, and carrying out quality check and standardization and formatting of output data; And 5, performing systematic data management. The step 1 specifically comprises the following steps: The method comprises the steps of 1.1, establishing a multi-product commodity template library, generating printing by utilizing a parameterization control mode through a programmed synthesis engine based on the commodity template library, and enriching and generating printing patterns, layout and background of the printing by adopting a random combination strategy and a diversity control m