Research map / Architectures and efficiency / CNN design
Dynamic Spatial Propagation Network for Depth Completion
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace fixed CSPN propagation with dynamic attention-based DySPN (ablation) | 7×7 CSPN · KITTI DC validation | RMSE (mm): 811.0 -> 769.0 | Helped |
| Increase DySPN iterations from 6 to 12 (ablation) | 7×7 DySPN · KITTI DC validation | RMSE (mm): 769.0 -> 768.1 | No clear effect |
| Use softmax instead of sigmoid attention (ablation) | 7×7 DySPN · KITTI DC validation | RMSE (mm): 769.0 -> 770.0 | No clear effect |
| Add diffusion suppression (ablation) | 7×7 DySPN · KITTI DC validation | RMSE (mm): 769.0 -> 763.9 | Helped |
| Add weight decay 5×10^5 (ablation) | 7×7 DySPN with DS · KITTI DC validation | RMSE (mm): 763.9 -> 753.2 | Helped |
| Add weight decay 10^5 (ablation) | 7×7 DySPN with DS · KITTI DC validation | RMSE (mm): 763.9 -> 752.9 | Helped |
| Add weight decay 5×10^6 (ablation) | 7×7 DySPN with DS · KITTI DC validation | RMSE (mm): 763.9 -> 755.8 | Helped |
| Add stochastic depth without weight decay (ablation) | 7×7 DySPN with DS · KITTI DC validation | RMSE (mm): 763.9 -> 745.8 | Helped |
| Use stochastic depth together with weight decay 10^5 instead of stochastic depth alone (ablation) | 7×7 DySPN with DS · KITTI DC validation | RMSE (mm): 745.8 -> 747.0 | Hurt |
| Use dilated neighborhoods (ablation) | 7×7 DySPN · KITTI DC validation | RMSE (mm): 757.2 -> 748.5 | Helped |
| Use deformable neighborhoods (ablation) | 7×7 DySPN · KITTI DC validation | RMSE (mm): 757.2 -> 739.4 | Helped |
| Use dilated neighborhoods (ablation) | 7×7 DySPN · KITTI DC validation | runtime (ms): 38 -> 6.9 | Helped |
Closest papers by meaning
- Bilateral Propagation Network for Depth Completion (CVPR 2024 · similarity 0.90)
- Masked Spatial Propagation Network for Sparsity-Adaptive Depth Refinement (CVPR 2024 · similarity 0.87)
- DeCoTR: Enhancing Depth Completion with 2D and 3D Attentions (CVPR 2024 · similarity 0.87)
- Completion as Enhancement: A Degradation-Aware Selective Image Guided Network for Depth Completion (CVPR 2025 · similarity 0.86)
- Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior (AAAI 2025 · similarity 0.85)
- Flexible Depth Completion for Sparse and Varying Point Densities (CVPR 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.