Dual-Consistency Representation Learning for Class-Incremental Semantic Segmentation in Remote Sensing Images (opens in new tab)

The class-incremental semantic segmentation (CISS) in remote sensing images focuses on enabling models to progressively adapt to new object categories without losing the ability to accurately segment previously learned ones. However, existing CISS frameworks often struggle with catastrophic forgetting (CF) of old classes and insufficient feature discrimination for new classes. To mitigate these issues, we introduce a dual-consistency representation learning (DCRL) framework for CISS in remote...