Abstract: Highlights•A boundary-aware dynamic re-weighting network (BDRN) for class-imbalanced SSL.•Shape constraint enhances boundary feature learning and mitigates confirmation bias.•Conflict-adversarial module ensures consistent segmentation across different views.•Dynamic re-weighting strategy improve segmentation performance on minority classes.
External IDs:doi:10.1016/j.eswa.2025.128175
Loading