Research map / 3D vision / Depth and stereo
A Dark Flash Normal Camera
Zhihao Xia, Jason Lawrence, Supreeth Achar
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 NIR dark flash input (ablation) | Full method · Normal estimation ablation (5 lighting conditions) | mean angular error: 19.82 -> 12.14 (overexposure); 13.13 -> 12.08 (well lit) | Improved |
| add NIR input (ablation) | Full method · Normal estimation ablation (5 lighting conditions) | mean angular error under low light: 19.39 -> 12.10 | Improved |
| add RGB input (ablation) | Full method · Normal estimation ablation (5 lighting conditions) | mean angular error: 12.54 -> 12.08 (well lit) | Improved |
| add stereo loss (ablation) | Full method · Normal estimation ablation (5 lighting conditions) | mean angular error (well lit): 12.80 -> 12.08 | Improved |
| add photometric loss (ablation) | Full method · Normal estimation ablation (5 lighting conditions) | mean angular error (well lit): 12.77 -> 12.08 | Improved |
| add NIR photometric loss on top of RGB photometric loss (ablation) | Full method · Normal estimation ablation (5 lighting conditions) | mean angular error (well lit): 12.64 -> 12.08 | Improved |
| add Blinn-Phong specular term to Lambertian model (ablation) | Full method · Normal estimation ablation (5 lighting conditions) | mean angular error (well lit): 12.44 -> 12.08 | Improved |
| add Blinn-Phong specular term (ablation) | Full method · Normal estimation ablation (5 lighting conditions) | diffuse albedo quality | Improved |
| use RGB+NIR network instead of RGB-only network under overexposure | RGB-only network · Normal estimation ablation (5 lighting conditions) | mean angular error: 19.82 -> 12.14 | Improved |
| use our method instead of Nestmeyer et al. (well lit) | Nestmeyer et al. · Normal estimation vs prior methods | mean angular error: 14.82 -> 12.29 | Improved |
| use our method instead of Nestmeyer et al. (low light) | Nestmeyer et al. · Normal estimation vs prior methods | mean angular error: 25.56 -> 12.33 | Improved |
| use our method instead of SfSNet (shadows) | SfSNet · Normal estimation vs prior methods | mean angular error: 18.32 -> 12.27 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SfSNet — SfSNet: Learning Shape, Reflectance and Illuminance of Faces `in the Wild' (CVPR 2018) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DeepPS2: Revisiting Photometric Stereo Using Two Differently Illuminated Images (ECCV 2022 · similarity 0.85)
- Deep near-light photometric stereo for spatially varying reflectances (ECCV 2020 · similarity 0.85)
- Stereoscopic Flash and No-Flash Photography for Shape and Albedo Recovery (CVPR 2020 · similarity 0.85)
- SPLiT: Single Portrait Lighting Estimation via a Tetrad of Face Intrinsics (TPAMI 2023 · similarity 0.85)
- Monocular Normal Estimation via Shading Sequence Estimation (ICLR 2026 · similarity 0.85)
- Faces as Lighting Probes via Unsupervised Deep Highlight Extraction (ECCV 2018 · 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.
