Symbiosis-Inspired Knowledge Distillation for Incremental Object Detection
July 16, 2026
The SIKD method mitigates catastrophic forgetting in incremental object detection by leveraging spatial and semantic dependencies between objects. By utilizing Spatial Symbiosis Distillation, the model preserves shared representations in regions where old and new classes overlap.
HOW THIS AFFECTS YOU
●
researcherYou can improve incremental learning performance by explicitly modeling object co-occurrence and occlusion dependencies.