Research map / 3D vision / Depth and stereo
Scale Propagation Network for Generalizable Depth Completion
Haotian Wang, Meng Yang, Xinhu Zheng, Gang Hua
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 |
|---|---|---|---|
| replace conventional BN/IN/LN with SP-Norm (ablation) | SPNet Tiny · Sintel | Rel: 0.448 -> 0.137 | Improved |
| use SP-Norm instead of ReZero (ablation) | SPNet Tiny · Sintel | Rel: 0.773 -> 0.137 | Improved |
| add SP-Norm in place of LN in the basic block (ablation) | SPNet · Sintel | Rel: 0.448 -> 0.137 | Improved |
| remove GRN from basic block (ablation) | SPNet · Sintel | Rel: 0.541 -> 0.137 | Improved |
| add data augmentation (Resized-Crop, Depth-Scaling) (ablation) | SPNet · Sintel | Rel: 0.554 -> 0.137 | Improved |
| use SLP instead of affine factors in SP-Norm (ablation) | SP-Norm · Sintel | Rel: 0.281 -> 0.137 | Improved |
| keep normalization operator in SP-Norm (ablation) | SP-Norm · all six test sets | Rel: convergence failure without it | Improved |
| use multiplier instead of adder in SP-Norm (ablation) | SP-Norm · all six test sets | Rel: convergence failure with adder | Improved |
| replace GELU, GRN and original block with modified block (ablation) | SPNet Tiny · 3090 GPU speed | speed: 114.1 -> 126.6 image/s | Improved |
| apply SP-Norm to DFU | DFU · KITTI | Rel: 0.485 -> 0.148 | Improved |
| apply SP-Norm to NLSPN | NLSPN · Sintel | Rel: 47.151 -> 49.778 | Hurt |
| apply SP-Norm to CFormer | CFormer · Sintel | Rel: 33.687 -> 37.247 | Hurt |
Closest papers by meaning
- Deep Depth Completion from Extremely Sparse Data: A Survey (TPAMI 2022 · similarity 0.91)
- CSPN++: Learning Context and Resource Aware Convolutional Spatial Propagation Networks for Depth Completion (AAAI 2020 · similarity 0.90)
- Bilateral Propagation Network for Depth Completion (CVPR 2024 · similarity 0.89)
- Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior (AAAI 2025 · similarity 0.89)
- 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 0.88)
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.
