Research map / Detection and segmentation / CNN features
Pattern-Affinitive Propagation Across Depth, Surface Normal and Semantic Segmentation
Zhenyu Zhang, Zhen Cui, Chunyan Xu, Yan Yan, Nicu Sebe, Jian Yang
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 |
|---|---|---|---|
| Jointly predict depth and normal with PAP rather than depth alone | NYUD-v2 depth estimation · NYUD-v2 | rmse: 0.570 -> 0.550 | Improved |
| Add cross-task propagation (ablation) | PAP joint depth, segmentation, and normal model · NYUD-v2 | rmse: 0.574 -> 0.558 | Improved |
| Add the reconstruction network (ablation) | PAP with cross-task propagation · NYUD-v2 | rmse: 0.558 -> 0.550 | Improved |
| Add pair-wise loss (ablation) | PAP with cross-task propagation and reconstruction network · NYUD-v2 | rmse: 0.550 -> 0.543 | Improved |
| Learn the affinity matrix at a larger scale, from 1/16 to 1/8 input scale (ablation) | PAP · NYUD-v2 | rmse: 0.543 -> 0.533 | Improved |
| Increase affinity-matrix scale from 1/8 to 1/4 input scale (ablation) | PAP · NYUD-v2 | rmse: 0.533 -> 0.530 | Improved |
| Use L1 distance rather than inner product for similarity calculation (ablation) | PAP · NYUD-v2 | rmse: 0.543 -> 0.540 | No clear effect |
| Jointly learn depth, segmentation, and normal rather than depth alone (ablation) | PAP · NYUD-v2 | rmse: 0.570 -> 0.533 | Improved |
| Use PAP for depth estimation | NYUD-v2 depth estimation · NYUD-v2 | rmse: 0.509 -> 0.497 | Improved |
| Use PAP for semantic segmentation | NYUD-v2 semantic segmentation · NYUD-v2 | IoU: 50.2 -> 50.4 | Improved |
| Use PAP for semantic segmentation | SUN-RGBD semantic segmentation · SUN-RGBD | IoU: 50.3 -> 50.5 | Improved |
| Apply PAP rather than single-task prediction | KITTI depth estimation · KITTI online benchmark | SILog: 14.58 -> 13.08 | Hurt |
Closest papers by meaning
- Pattern-Structure Diffusion for Multi-Task Learning (CVPR 2020 · similarity 0.88)
- Visual Micro-Pattern Propagation (TPAMI 2022 · similarity 0.86)
- Learning Affinity via Spatial Propagation Networks (NeurIPS 2017 · similarity 0.86)
- Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2025 · similarity 0.84)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (CVPR 2024 · similarity 0.84)
- UniDSeg: Unified Cross-Domain 3D Semantic Segmentation via Visual Foundation Models Prior (NeurIPS 2024 · similarity 0.84)
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.
