Research map / 3D vision / Face and shape
Towards Racially Unbiased Skin Tone Estimation via Scene Disambiguation
Haiwen Feng, Timo Bolkart, Joachim Tesch, Michael J. Black, Victoria Abrevaya
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 synthetic-data semi-supervised training instead of self-supervised only (ablation) | faceSH · FAIR benchmark | total score: 35.62 -> 21.31 | Improved |
| add scene consistency loss (ablation) | fuseSH · FAIR benchmark | bias score: 11.08 -> 3.50 | Improved |
| add scene consistency loss (ablation) | fuseSH · FAIR benchmark | avg ITA error: 15.22 -> 15.82 | No clear effect |
| add scene consistency loss (ablation) | fuseSH · FAIR benchmark | MAE: 17.02 -> 20.37 | Hurt |
| add light-intensity conditioning of albedo encoder (ablation) | fuseSH · FAIR benchmark | avg ITA error: 15.22 -> 14.16 | Improved |
| add light-intensity conditioning of albedo encoder (ablation) | fuseSH · FAIR benchmark | bias score: 11.08 -> 6.56 | Improved |
| add conditioning to fuseSH + sc (full model) (ablation) | fuseSH + sc · FAIR benchmark | total score: 19.32 -> 16.81 | Improved |
| add scene consistency to fuseSH + cond (full model) (ablation) | fuseSH + cond · FAIR benchmark | bias score: 6.56 -> 2.63 | Improved |
| use scene-level light intensity (fuseSH) instead of crop-only light, semi-supervised (ablation) | faceSH + semi · FAIR benchmark | total score: 21.31 -> 26.31 | Hurt |
| use balanced albedo model instead of BFM albedo space (ablation) | TRUST · FAIR benchmark | avg ITA error: 16.19 -> 13.87 | Improved |
| use balanced albedo model instead of BFM albedo space (ablation) | TRUST · FAIR benchmark | bias score: 15.33 -> 2.79 | Improved |
| use balanced albedo model instead of AlbedoMM (ablation) | TRUST · FAIR benchmark | total score: 33.00 -> 16.67 | 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.
- GANFIT — GANFIT: Generative Adversarial Network Fitting for High Fidelity 3D Face Reconstruction (CVPR 2019) · better on 3 of 3 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
- Improving Fairness in Facial Albedo Estimation via Visual-Textual Cues (CVPR 2023 · similarity 0.93)
- HUST: High-Fidelity Unbiased Skin Tone Estimation via Texture Quantization (ICCV 2025 · similarity 0.89)
- S^3-Face: SSS-Compliant Facial Reflectance Estimation via Diffusion Priors (CVPR 2025 · similarity 0.87)
- Monocular Identity-Conditioned Facial Reflectance Reconstruction (CVPR 2024 · similarity 0.87)
- UltrAvatar: A Realistic Animatable 3D Avatar Diffusion Model with Authenticity Guided Textures (CVPR 2024 · similarity 0.85)
- MoSAR: Monocular Semi-Supervised Model for Avatar Reconstruction using Differentiable Shading (CVPR 2024 · 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.
