Research map / Detection and segmentation / CNN features
Bi-directional Cross-Modality Feature Propagation with Separation-and-Aggregation Gate for RGB-D Semantic Segmentation
Xiaokang Chen, Kwan-Yee Lin, Jingbo Wang, Wayne Wu, Chen Qian, Hongsheng Li, Gang Zeng
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 |
|---|---|---|---|
| add SA-Gate to the dual-path ResNet-50 (ablation) | Res50 dual path · NYU Depth V2 test (no decoder) | mIoU: 45.9 -> 47.4 | Improved |
| add BMP to the dual-path ResNet-50 (ablation) | Res50 dual path · NYU Depth V2 test (no decoder) | mIoU: 45.9 -> 47.8 | Improved |
| add both BMP and SA-Gate (ablation) | Res50 dual path · NYU Depth V2 test (no decoder) | mIoU: 45.9 -> 48.6 | Improved |
| use proposed cross-modality guided FS instead of product re-weighting (ablation) | SA-Gate · NYU Depth V2 test (no decoder) | mIoU: 47.5 -> 48.6 | Improved |
| use proposed FS instead of self-global filtering (ablation) | SA-Gate · NYU Depth V2 test (no decoder) | mIoU: 47.5 -> 48.6 | Improved |
| use proposed FA gating instead of addition (ablation) | SA-Gate · NYU Depth V2 test (no decoder) | mIoU: 47.8 -> 48.6 | Improved |
| use proposed FA gating instead of conv on concatenation (ablation) | SA-Gate · NYU Depth V2 test (no decoder) | mIoU: 48.0 -> 48.6 | Improved |
| place SA-Gate in block 1 instead of block 4 (ablation) | Res50* · NYU Depth V2 test (no decoder) | mIoU: 44.3 -> 47.8 | Improved |
| stack SA-Gate on blocks 1-3 rather than block 1 alone (ablation) | Res50* · NYU Depth V2 test (no decoder) | mIoU: 47.8 -> 48.3 | Improved |
| add SA-Gate at block 4 on top of blocks 1-3 (ablation) | Res50* · NYU Depth V2 test (no decoder) | mIoU: 48.3 -> 48.0 | Hurt |
| replace parallel DeepLab V3+ RGB-D baseline with the proposed encoder | DeepLab V3+ · NYU Depth V2 test | mIoU: 46.7 -> 50.4 | Improved |
| replace RGB-D baseline with the proposed encoder | DeepLab V3+ RGB-D · CityScapes val | mIoU: 80.5 -> 81.7 | Improved |
Closest papers by meaning
- Keep the Balance: A Parameter-Efficient Symmetrical Framework for RGB+X Semantic Segmentation (CVPR 2025 · similarity 0.89)
- DFormer: Rethinking RGBD Representation Learning for Semantic Segmentation (ICLR 2024 · similarity 0.88)
- DFormerv2: Geometry Self-Attention for RGBD Semantic Segmentation (CVPR 2025 · similarity 0.87)
- 3D Vision-Language Gaussian Splatting (ICLR 2025 · similarity 0.86)
- PDDM: Pseudo Depth Diffusion Model for RGB-PD Semantic Segmentation Based in Complex Indoor Scenes (AAAI 2025 · similarity 0.86)
- Re-coding for Uncertainties: Edge-awareness Semantic Concordance for Resilient Event-RGB Segmentation (NeurIPS 2025 · similarity 0.86)
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.
