CroDiNo-KD jointly trains RGB and depth models for semantic segmentation using disentanglement and contrastive losses, beating teacher-based cross-modal distillation on three benchmarks.
Class attention transfer based knowledge distillation,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Revisiting Cross-Modal Knowledge Distillation: A Disentanglement Approach for RGBD Semantic Segmentation
CroDiNo-KD jointly trains RGB and depth models for semantic segmentation using disentanglement and contrastive losses, beating teacher-based cross-modal distillation on three benchmarks.