Research map / 3D vision / Depth and stereo
Normal Assisted Stereo Depth Estimation
Check 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 |
|---|---|---|---|
| jointly estimate normals from the multi-view cost volume alongside depth | DPSNet stereo depth estimation · SUN3D | Abs Rel: 0.1470 -> 0.1332 | Helped |
| jointly estimate normals from the multi-view cost volume alongside depth | DPSNet stereo depth estimation · Scene Flow | EPE: 0.80 -> 0.69 | Helped |
| add pixel-coordinate depth-normal consistency refinement (ablation) | joint depth-normal model · ScanNet | Abs Rel: 0.1150 -> 0.1070 | Helped |
| add pixel-coordinate depth-normal consistency refinement (ablation) | joint depth-normal model · SUN3D | Abs Rel: 0.1332 -> 0.1247 | Helped |
| replace pixel-coordinate consistency with world-coordinate consistency loss (ablation) | UNet depth refinement · SUN3D | Abs Rel: 0.1247 -> 0.1288 | Hurt |
| use UNet refinement with ground-truth supervision only, without consistency loss (ablation) | joint depth-normal model · SUN3D | Abs Rel: 0.1247 -> 0.1307 | Hurt |
| remove the refinement UNet (ablation) | joint depth-normal model · SUN3D | Abs Rel: 0.1247 -> 0.1332 | Hurt |
| add normal supervision to the cost volume | stereo cost-volume depth estimation · DeMoN datasets without cost aggregation | Abs Rel: 0.1274 -> 0.1114 | Helped |
| add a normal-estimation branch to GANet | GANet · Scene Flow | EPE: 0.78 -> 0.77 | Helped |
| use multi-view cost-volume normal estimation instead of single-view RGB normal estimation | normal estimation · ScanNet | Mean angle error: 31.1 -> 23.1 | Helped |
| use multi-view cost-volume normal estimation instead of single-view RGB estimation | normal estimation trained on ScanNet · SUN3D | Mean angle error: 31.6 -> 22.9 | Helped |
Closest papers by meaning
- Rethinking Depth Estimation for Multi-View Stereo: A Unified Representation (CVPR 2022 · similarity 0.90)
- Just Look at the Image: Viewpoint-Specific Surface Normal Prediction for Improved Multi-View Reconstruction (CVPR 2016 · similarity 0.90)
- Single View Stereo Matching (CVPR 2018 · similarity 0.89)
- MVSAnywhere: Zero-Shot Multi-View Stereo (CVPR 2025 · similarity 0.88)
- RNb-NeuS: Reflectance and Normal-based Multi-View 3D Reconstruction (CVPR 2024 · similarity 0.87)
- Leveraging Global Stereo Consistency for Category-Level Shape and 6D Pose Estimation from Stereo Images (CVPR 2025 · similarity 0.87)
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.