CTFS: Collaborative Teacher Framework for Forward-Looking Sonar Image Semantic Segmentation with Extremely Limited Labels
The CTFS framework solves poor semantic segmentation performance for forward-looking sonar images when labeled data is scarce by using a collaborative teacher system and a reliability check. It combines one general teacher with several sonar-specific teachers to teach the model both general features and unique sonar characteristics. A cross-teacher assessment mechanism filters noisy pseudo-labels, resulting in a 5.08% mIoU improvement on the FLSMD dataset with only 2% labeled data.