Graph Convolutional Network for Image Segmentation Annotation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image region segmentation methods using weak supervision with image-level labels are inadequate for precise segmentation due to insufficient supervision, leading to inaccurate results, especially in time-consuming and labor-intensive tasks like medical image analysis.
Innovation Solution
The method involves converting image-level annotation information into superpixel-level annotation information using a graph convolutional network model, enabling stronger supervision during model training and improving segmentation precision by utilizing graph structure data with nodes representing pixels or superpixels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If weak supervision with image-level labels is used for image region segmentation, then annotation time and labor costs are reduced, but segmentation precision deteriorates
Solution Approach 1:
The patent introduces superpixel-level annotation information as an intermediary between image-level labels and pixel-level segmentation. The graph convolutional network model generates this intermediate representation by propagating image-level annotations to superpixel regions, providing stronger supervision signals without requiring manual pixel-level annotation, thus resolving the contradiction between annotation efficiency and segmentation precision
Solution Approach 2:
The patent transforms the annotation dimension from image-level (coarse) to superpixel-level (fine) by introducing a new level of granularity. This dimensional transformation allows the model to leverage graph structure data and spatial relationships among superpixels, enhancing segmentation precision while maintaining the efficiency of image-level annotation
2Measurement precision
If pixel-level annotation is used for training, then segmentation precision is improved, but annotation time and labor costs increase significantly
Solution Approach 1:
The patent creates pseudo pixel-level annotation information by copying and propagating image-level labels through the graph convolutional network model to superpixel regions. This generated annotation data serves as a substitute for manual pixel-level annotation, providing sufficient supervision signals for precise segmentation while avoiding the time-consuming manual annotation process
Solution Approach 2:
The system performs self-annotation by automatically generating superpixel-level annotation information from image-level labels using the graph convolutional network model. This self-service mechanism eliminates the need for external manual pixel-level annotation while still providing strong supervision signals for training the segmentation model
Data Source
AI summary
Embodiments of this application disclose a method and apparatus for training an image region segmentation model, and an image region segmentation method and apparatus. The method includes acquiring a sample image set, and each image of the sample image set having first annotation information; generating graph structure data corresponding to a sample image in the sample image set, the graph structure data comprising multiple nodes, and each node comprising at least one pixel in the sample image; determining second annotation information of each node according to the graph structure data and the first annotation information corresponding to the sample image by using a graph convolutional network model, a granularity of the second annotation information being smaller than a granularity of the first annotation information, the graph convolutional network model being a part of an image segmentation model; and training the image segmentation model according to the second annotation information.