Research map / Detection and segmentation / CNN features
Learning Common and Specific Features for RGB-D Semantic Segmentation with Deconvolutional Networks
Jinghua Wang, Zhenhua Wang, Dacheng Tao, Simon See, Gang Wang
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 MK-MMD-regularized common and specific feature learning over the unregularized network (ablation) | Ours · NYU Depth V1 (13-class) | class-average accuracy (%): 69.9% -> 78.8% | Improved |
| add regularization over the unregularized network (ablation) | Ours · NYU Depth V2 (13-class) | class-average accuracy (%): 49.2% -> 52.7% | Improved |
| add regularization over the unregularized network (ablation) | Ours · NYU Depth V2 (40-class) | class-average accuracy (%): 41.7% -> 47.3% | Improved |
| use MK-MMD instead of Euclidean distance for feature similarity (ablation) | Ours · NYU Depth V1 (13-class) | class-average accuracy (%): 71.4% -> 78.8% | Improved |
| use MK-MMD instead of Euclidean distance for feature similarity (ablation) | Ours · NYU Depth V2 (13-class) | class-average accuracy (%): 46.1 -> 52.7 | Improved |
| learn common and specific features jointly instead of fusing independent per-modality networks at decision level (ablation) | Ours · NYU Depth V1 (13-class) | class-average accuracy (%): 76.5% -> 78.8% | Improved |
| learn common and specific features jointly instead of fusing independent per-modality networks at decision level (ablation) | Ours · NYU Depth V2 (13-class) | class-average accuracy (%): 47.6 -> 52.7 | Improved |
| separate common and specific features instead of one deconvolutional network on concatenated features (S-DN) (ablation) | Ours · NYU Depth V2 (13-class) | class-average accuracy (%): 45.6 -> 52.7 | Improved |
| disentangle common and specific features instead of four-channel RGB-D input (C-DN) (ablation) | Ours · NYU Depth V2 (13-class) | class-average accuracy (%): 45.1 -> 52.7 | Improved |
| use the proposed method over the previous state of the art on 13 classes | Ours · NYU Depth V2 (13-class) | class-average accuracy (%): 48.0% -> 52.7% | Improved |
| per-class comparison: proposed method on bed against Couprie | Ours · NYU Depth V2 (13-class), bed | per-class accuracy (%): 38.1 -> 31.6 | Hurt |
Closest papers by meaning
- Locality-Sensitive Deconvolution Networks With Gated Fusion for RGB-D Indoor Semantic Segmentation (CVPR 2017 · similarity 0.93)
- RDFNet: RGB-D Multi-Level Residual Feature Fusion for Indoor Semantic Segmentation (ICCV 2017 · similarity 0.92)
- UCTNet: Uncertainty-Aware Cross-Modal Transformer Network for Indoor RGB-D Semantic Segmentation (ECCV 2022 · similarity 0.89)
- ShapeConv: Shape-Aware Convolutional Layer for Indoor RGB-D Semantic Segmentation (ICCV 2021 · similarity 0.88)
- DFormer: Rethinking RGBD Representation Learning for Semantic Segmentation (ICLR 2024 · similarity 0.86)
- Geometric Exploitation for Indoor Panoramic Semantic Segmentation (NeurIPS 2024 · similarity 0.85)
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.
