Research map / Detection and segmentation / Open-set and few-shot detection
Incremental Class Discovery for Semantic Segmentation With RGBD Sensing
Yoshikatsu Nakajima, Byeongkeun Kang, Hideo Saito, Kris Kitani
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| combine geometric and deep features with entropy-based confidence weighting instead of geometric features only (ablation) | Ours · NYUDv2 (140 sequences, 13 classes) | mean IoU: 42.24 -> 46.05 | Improved |
| combine geometric and deep features with entropy-based confidence weighting instead of deep features only (ablation) | Ours · NYUDv2 (140 sequences, 13 classes) | mean IoU: 42.22 -> 46.05 | Improved |
| use RGBD SLIC-based incremental segmentation (color + depth) instead of depth-only 3D segmentation | Ours · NYUDv2 (140 sequences, 13 classes) | mean IoU: 41.59 -> 46.05 | Improved |
| use deep features only instead of geometric only (trained classes) (ablation) | Ours · NYUDv2 (140 sequences, 13 classes) trained classes | mean IoU: bed 51.95 -> 60.07 | Mixed |
| use geometric features instead of deep features only on unseen classes (ablation) | Ours · NYUDv2 (140 sequences, 13 classes) novel classes | mean IoU: window 42.30 -> 52.43 | Improved |
| add color via SLIC segmentation on poor-geometry classes (picture) | Ours · NYUDv2 (140 sequences, 13 classes) picture | mean IoU: 19.37 -> 28.69 | Improved |
| add color via SLIC segmentation on poor-geometry classes (window) | Ours · NYUDv2 (140 sequences, 13 classes) window | mean IoU: 31.30 -> 53.92 | Improved |
| cluster segments instead of elements (processing time of clustering) | Ours · NYUDv2 (140 sequences, 13 classes) | processing time: 13.4 ms | Improved |
| store features per segment instead of per element (memory) | Ours · NYUDv2 (140 sequences, 13 classes) | memory footprint (number of stored units): 196 vs 900,478 | Improved |
Closest papers by meaning
- PointPDF V2: A Unified Framework for Continual Open-World 3D Semantic Segmentation (TPAMI 2026 · similarity 0.89)
- DFormer: Rethinking RGBD Representation Learning for Semantic Segmentation (ICLR 2024 · similarity 0.88)
- LangRef3DGS: Natural Language-Guided 3D Referential Segmentation from Partial Observations via 3D Gaussian Splatting (CVPR 2026 · similarity 0.88)
- Cs2K: Class-specific and Class-shared Knowledge Guidance for Incremental Semantic Segmentation (ECCV 2024 · similarity 0.88)
- 3D-SIS: 3D Semantic Instance Segmentation of RGB-D Scans (CVPR 2019 · similarity 0.88)
- Open-World Semantic Segmentation Including Class Similarity (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
