Semantic Segmentation Model Training via Self-Supervised Sub-Image Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image semantic segmentation technologies face challenges in achieving high accuracy and efficiency, particularly in self-supervised learning scenarios where labeled data is limited, leading to suboptimal performance in distinguishing semantic categories.
Innovation Solution
The proposed method involves a semantic segmentation model that performs image semantic segmentation on unlabeled images, extracts sub-images using a convolutional neural network based on category and feature distances, and trains the model using a self-supervised approach to enhance semantic distinction and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If self-supervised learning is used with limited labeled data, then training efficiency is improved, but segmentation accuracy deteriorates
Solution Approach 1:
The method performs preliminary semantic segmentation on unlabeled images to obtain preliminary segmentation results before using them as training data. This preliminary action enables the model to learn from unlabeled data while maintaining accuracy by subsequently refining these results with labeled data in a two-stage training process
Solution Approach 2:
The model performs self-supervised learning by using its own preliminary segmentation results as training targets. The system generates its own training data from unlabeled images through self-segmentation, reducing dependency on externally labeled data while maintaining learning effectiveness
2Measurement precision
If more labeled data is used for training, then segmentation accuracy is improved, but data preparation complexity increases
Solution Approach 1:
The system generates its own training data by performing self-segmentation on unlabeled images. This self-service approach eliminates the need for extensive manual labeling, reducing data preparation complexity while still providing sufficient training data for accurate segmentation through the self-generated preliminary results
Solution Approach 2:
The method performs preliminary segmentation on unlabeled data before formal training, creating a foundation that reduces the amount of labeled data needed. This preliminary action on unlabeled data decreases data preparation complexity while maintaining accuracy benefits
Data Source
AI summary
A semantic segmentation model training method includes: performing, by a semantic segmentation model, image semantic segmentation on at least one unlabeled image to obtain a preliminary semantic segmentation result as the category of the unlabeled image; obtaining, by a convolutional neural network based on the category of the at least one unlabeled image and the category of at least one labeled image, sub-images respectively corresponding to the at least two images and features corresponding to the sub-images, where the at least two images comprise the at least one unlabeled image and the at least one labeled image, and the at least two sub-images carry the categories of the corresponding images; and training the semantic segmentation model on the basis of the categories of the at least two sub-images and feature distances between the at least two sub-images.