Research map / 3D vision / Depth and stereo
CSPN++: Learning Context and Resource Aware Convolutional Spatial Propagation Networks 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 |
|---|---|---|---|
| Use CA-CSPN instead of the strongest listed baseline | KITTI Depth Completion benchmark · KITTI Depth Completion benchmark | RMSE (mm): 752.88 -> 743.69 | Helped |
| Add confidence-guided replacement so sparse depths are preserved according to predicted confidence (ablation) | CSPN · KITTI validation | RMSE (mm): 765.78 -> 756.27 | Helped |
| Add learned kernel assembly (ablation) | CSPN with guided replacement · KITTI validation | RMSE (mm): 756.27 -> 732.46 | Helped |
| Add learned iteration assembly (ablation) | Kernel-assembled CA-CSPN · KITTI validation | RMSE (mm): 732.46 -> 732.34 | Helped |
| Add latency regularization (ablation) | CA-CSPN with kernel and iteration assembly · KITTI validation | RMSE (mm): 732.34 -> 725.43 | Helped |
| Add SPP to the baseline (ablation) | Depth completion baseline · KITTI validation | RMSE (mm): 799.08 -> 788.23 | Helped |
| Add CSPN refinement (ablation) | SPP baseline · KITTI validation | RMSE (mm): 788.23 -> 765.78 | Helped |
| Use RA-CSPN configuration selection instead of fixed CSPN configuration (ablation) | CSPN · KITTI | RMSE (mm): 756.27 -> 732.32 | Helped |
| Select per-pixel configurations with RA-CSPN rather than use the fixed CSPN setup (ablation) | CSPN · KITTI | latency (ms): 28.88 -> 10.03 | Helped |
| Impose memory and latency budgets of 0.35 (ablation) | RA-CSPN · KITTI | RMSE (mm): 732.32 -> 742.17 | Hurt |
| Impose memory and latency budgets of 0.40 (ablation) | RA-CSPN · NYU v2 | RMSE (mm): 121.49 -> 118.70 | Helped |
Closest papers by meaning
- Bilateral Propagation Network for Depth Completion (CVPR 2024 · similarity 0.90)
- Completion as Enhancement: A Degradation-Aware Selective Image Guided Network for Depth Completion (CVPR 2025 · similarity 0.88)
- DeCoTR: Enhancing Depth Completion with 2D and 3D Attentions (CVPR 2024 · similarity 0.88)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.87)
- Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior (AAAI 2025 · similarity 0.86)
- Masked Spatial Propagation Network for Sparsity-Adaptive Depth Refinement (CVPR 2024 · 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.