Research map / 3D vision / Depth and stereo
Faces as Lighting Probes via Unsupervised Deep Highlight Extraction
Renjiao Yi, Chenyang Zhu, Ping Tan, Stephen Lin
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 unsupervised low-rank finetuning on real celebrity face images after synthetic pretraining (ablation) | Highlight-Net · highlight removal on lab faces | RMSE / SSIM | Improved |
| use finetuned Highlight-Net instead of the strongest prior highlight removal | Highlight-Net · synthetic faces | mean RMSE: 4.15 -> 3.37 | Improved |
| use finetuned Highlight-Net instead of the strongest prior highlight removal | Highlight-Net · real faces | mean RMSE: 8.93 -> 7.61 | Improved |
| use finetuned Highlight-Net instead of the strongest prior highlight removal | Highlight-Net · real faces | mean SSIM: 0.89 -> 0.89 | No clear effect |
| use face-based non-parametric environment map instead of best prior illumination method | face lighting probe · diffuse bunny relighting, outdoor | relighting RMSE: 17.77 -> 10.78 | Improved |
| use face-based non-parametric environment map instead of best prior illumination method | face lighting probe · diffuse bunny relighting, indoor | relighting RMSE: 20.52 -> 13.18 | Improved |
| use face-based non-parametric environment map instead of best prior illumination method | face lighting probe · glossy bunny relighting, indoor | relighting RMSE: 21.01 -> 13.69 | Improved |
| use face-based method instead of best prior on real captured scenes | face lighting probe · real scenes outdoor | normalized RMSE: 0.154 -> 0.143 | Improved |
| use face-based method instead of best prior on real captured scenes | face lighting probe · real scenes indoor | normalized RMSE: 0.050 -> 0.045 | Improved |
| use ground-truth face positions instead of estimated ones for light triangulation | face lighting probe · lamp 1 position | position error (m): 0.41 -> 0.20 | Improved |
| use ground-truth face positions instead of estimated ones for light triangulation | face lighting probe · lamp 2 position | position error (m): 0.51 -> 0.49 | No clear effect |
Closest papers by meaning
- SPLiT: Single Portrait Lighting Estimation via a Tetrad of Face Intrinsics (TPAMI 2023 · similarity 0.88)
- Learning Physics-Guided Face Relighting Under Directional Light (CVPR 2020 · similarity 0.88)
- Towards High Fidelity Face Relighting With Realistic Shadows (CVPR 2021 · similarity 0.87)
- Learning to Decouple the Lights for 3D Face Texture Modeling (NeurIPS 2024 · similarity 0.87)
- SfSNet: Learning Shape, Reflectance and Illuminance of Faces `in the Wild' (CVPR 2018 · similarity 0.87)
- SynthLight: Portrait Relighting with Diffusion Model by Learning to Re-render Synthetic Faces (CVPR 2025 · 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.
