Character Detector Training via Weak Supervision and Bounding Box Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current character detection methods based on character-level annotation require significant manpower and resources, and existing data sets annotated at the word or text bar level hinder effective training of character detectors, limiting their adaptability to real scenarios.
Innovation Solution
A method and apparatus for training a character detector using weak supervision, which inputs coarse-grained annotation information, divides it to obtain coarse bounding boxes, and adjusts them based on predicted bounding boxes to generate fine bounding boxes, enabling character-level annotation and improving detection precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If character-level annotation is used for training, then character detection precision is improved, but annotation cost and time consumption increase significantly
Solution Approach 1:
The patent applies preliminary action by performing coarse-grained annotation at the word or text bar level first, then automatically generating character-level bounding boxes through division and prediction. This preliminary coarse annotation avoids the need for direct manual character-level annotation, significantly reducing annotation time while still enabling character detection training.
Solution Approach 2:
The patent segments the annotation process into two stages: coarse-grained annotation (word/text bar level) and fine-grained generation (character level). By dividing the annotation task from coarse to fine, it reduces the manual annotation burden while maintaining character-level detection precision through automated refinement.
2Ease of manufacture
If word or text bar level annotation is used, then annotation difficulty and cost are reduced, but character detection training effectiveness deteriorates
Solution Approach 1:
The patent introduces an intermediary neural network model that bridges the gap between coarse-grained word/text bar annotations and fine-grained character-level annotations. This intermediary automatically generates character bounding boxes from coarse annotations, maintaining training effectiveness without requiring direct character-level annotation.
Solution Approach 2:
The patent performs preliminary coarse-grained annotation at the easier word or text bar level, then uses automated processing to generate the required character-level annotations. This preliminary action at a coarser level maintains ease of annotation while enabling effective character detection training through subsequent automated refinement.
3Productivity
If coarse-grained annotation is used for training, then annotation efficiency is improved, but character-level detection precision deteriorates
Solution Approach 1:
The patent segments the annotation process into coarse-grained (word/text bar level) and fine-grained (character level) stages. By dividing the annotation task, it maintains high annotation efficiency at the coarse level while automatically generating fine-grained character-level annotations through neural network prediction, thus preserving detection precision.
Solution Approach 2:
The patent performs preliminary coarse-grained annotation to improve efficiency, then applies automated refinement through neural network-based prediction to generate precise character-level bounding boxes. This two-stage approach maintains annotation efficiency while recovering character-level detection precision.
Data Source
AI summary
A method and apparatus for training a character detector based on weak supervision, a character detection system and a computer readable storage medium are provided, wherein the method includes: inputting coarse-grained annotation information of a to-be-processed object, wherein the coarse-grained annotation information including a whole bounding outline of a word, text bar or line of the to-be-processed object; dividing the whole bounding outline of the coarse-grained annotation information, to obtain a coarse bounding box of a character of the to-be-processed object; obtaining a predicted bounding box of the character of the to-be-processed object through a neural network model from the coarse-grained annotation information; and determining a fine bounding box of the character of the to-be-processed object as character-based annotation of the to-be-processed object, according to the coarse bounding box and the predicted bounding box.