Research map / Detection and segmentation / CNN features
Depth-aware CNN for RGB-D Segmentation
Weiyue Wang, Ulrich Neumann
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 depth-aware convolution to the lower layers (Conv1_1 to Conv3_1) in addition to upper layers (ablation) | D-CNN (VGG-16) · NYUv2 test (from scratch) | mIoU: 22.4 -> 27.8 | Improved |
| add depth-aware average pooling (ablation) | D-CNN (VGG-16) · NYUv2 test (from scratch) | mIoU: 27.4 -> 27.8 | Improved |
| replace standard convolution with depth-aware operations | DeepLab VGG-16 · NYUv2 test (from scratch) | mIoU: 15.9 -> 27.8 | Improved |
| use D-CNN instead of two-stream HHA network | DeepLab VGG-16 · NYUv2 test (from scratch) | mIoU: 21.9 -> 27.8 | Improved |
| replace standard convolution with depth-aware operations | DeepLab VGG-16 · SUN-RGBD test (from scratch) | mIoU: 22.8 -> 29.7 | Improved |
| replace standard convolution with depth-aware operations | DeepLab VGG-16 · SID Area 5 (from scratch) | mIoU: 35.5 -> 39.5 | Improved |
| add depth-aware convolution to ResNet-50 encoder | ResNet-50 · NYUv2 test (pretrained, ResNet-50) | mIoU: 38.8 -> 41.5 | Improved |
| add HHA encoding stream to D-CNN | D-CNN · NYUv2 test (pretrained) | mIoU: 41.0 -> 43.9 | Improved |
| add HHA encoding stream to D-CNN | D-CNN · NYUv2 test (from scratch) | mIoU: 27.8 -> 26.2 | Hurt |
| increase alpha in depth similarity from 8.3 to 20 (ablation) | D-CNN · NYUv2 test (from scratch) | mIoU: 27.8 -> 24.9 | Hurt |
| decrease alpha in depth similarity from 8.3 to 2.5 (ablation) | D-CNN · NYUv2 test (from scratch) | mIoU: 27.8 -> 25.3 | Hurt |
| use binary clipped depth similarity instead of exponential similarity (ablation) | D-CNN · NYUv2 test (from scratch) | mIoU: 27.8 -> 20.1 | Hurt |
Closest papers by meaning
- DFormerv2: Geometry Self-Attention for RGBD Semantic Segmentation (CVPR 2025 · similarity 0.90)
- ShapeConv: Shape-Aware Convolutional Layer for Indoor RGB-D Semantic Segmentation (ICCV 2021 · similarity 0.90)
- DFormer: Rethinking RGBD Representation Learning for Semantic Segmentation (ICLR 2024 · similarity 0.90)
- Depth CNNs for RGB-D Scene Recognition: Learning from Scratch Better than Transferring from RGB-CNNs (AAAI 2017 · similarity 0.90)
- Malleable 2.5D Convolution: Learning Receptive Fields along the Depth-axis for RGB-D Scene Parsing (ECCV 2020 · similarity 0.90)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (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.
