Research map / 3D vision / Depth and stereo
Surface Normals in the Wild
Weifeng Chen, Donglai Xiang, Jia Deng
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-based surface normal loss to relative depth training | d (relative depth only) · NYU Full metric depth | LS-RMSE: 0.52 -> 0.50 | Improved |
| add depth-based surface normal loss to relative depth training | d (relative depth only) · NYU Subset metric depth | LS-RMSE: 0.64 -> 0.58 | Improved |
| add angle-based surface normal loss to relative depth training | d_F · NYU Full metric depth | LS-RMSE: 0.52 -> 0.55 | Hurt |
| add angle-based surface normal loss to relative depth training | d · NYU Subset metric depth | LS-RMSE: 0.64 -> 0.65 | No clear effect |
| add multiscale normals to angle-based loss (ablation) | d_n_al · NYU Subset metric depth | LS-RMSE: 0.65 -> 0.59 | Improved |
| add multiscale normals to depth-based loss (ablation) | d_n_dl_F · NYU Full metric depth | LS-RMSE: 0.50 -> 0.52 | Hurt |
| add angle-based surface normal loss | d · NYU Subset ordinal error | WKDR: 37.6% -> 36.5% | Improved |
| add depth-based surface normal loss | d · NYU Subset ordinal error | WKDR: 37.6% -> 38.7% | Hurt |
| add multiscale normals to angle-based loss (ablation) | d_n_al · NYU Subset ordinal error | WKDR: 36.5% -> 34.6% | Improved |
| add angle-based surface normal loss | d_F · NYU Full derived normals | mean angle distance: 29.45 -> 25.92 | Improved |
| add angle-based surface normal loss | d · NYU Subset derived normals | mean angle distance: 45.46 -> 37.53 | Improved |
| fine-tune on SNOW | d_n_al_F · SNOW surface normals | mean angle distance: 32.53 -> 25.75 | Improved |
Closest papers by meaning
- Single-Image Depth Perception in the Wild (NeurIPS 2016 · similarity 0.90)
- GeoNet: Geometric Neural Network for Joint Depth and Surface Normal Estimation (CVPR 2018 · similarity 0.88)
- Deep Surface Normal Estimation on the 2-Sphere with Confidence Guided Semantic Attention (ECCV 2020 · similarity 0.87)
- Unsupervised Learning of Geometry From Videos With Edge-Aware Depth-Normal Consistency (AAAI 2018 · similarity 0.87)
- Hierarchical Normalization for Robust Monocular Depth Estimation (NeurIPS 2022 · similarity 0.87)
- Rethinking Inductive Biases for Surface Normal Estimation (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.
