Missing Is Not Zero: well-posed modality removal for RGB-D segmentation robustness. A removal-equivalence criterion with an inference-time test, plus 293 SHA-256-verified result artifacts the paper's headline table recomputes from.
computer-vision reproducible-research pytorch semantic-segmentation robustness multimodal nyu-depth-v2 rgbd-segmentation missing-modality dformer
-
Updated
Aug 1, 2026 - Python