Research map / 3D vision / Face and shape
Discovering Geometric Biases in 3D Face Reconstruction: A Curvature-Aware Spectral Framework for Fairness Evaluation
Veronika Shilova, Emmanuel Malherbe, Giovanni Palma, Panagiotis-Alexandros Bokaris, Laurent Risser, Jean-Michel Loubes
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 |
|---|---|---|---|
| use curvature reconstruction error instead of Euclidean NMSE (REALY mapping) | 3D face reconstruction evaluation · User study: pairwise agreement with human judgment (REALY nose, BFM, 1,111 pairs) | pairwise accuracy (%): 53.92 -> 73.63 | Improved |
| use L1 norm instead of L2 in CRE (ablation) | CRE · User study: pairwise agreement with human judgment (REALY nose, BFM, 1,111 pairs) | pairwise accuracy (%): 72.28 -> 73.63 | Improved |
| use L1 norm instead of Huber in CRE (ablation) | CRE · User study: pairwise agreement with human judgment (REALY nose, BFM, 1,111 pairs) | pairwise accuracy (%): 67.60 -> 73.63 | Improved |
| use smaller K (128 vs 1024 eigenfunctions) (ablation) | CRE (R=3, L1) · User study: pairwise agreement with human judgment (REALY nose, BFM, 1,111 pairs) | pairwise accuracy (%): 69.85 -> 73.63 | Improved |
| vary R from 1 to 4 at K=128 (ablation) | CRE · User study: pairwise agreement with human judgment (REALY nose, BFM, 1,111 pairs) | pairwise accuracy (%): 71.38 to 73.63 | No clear effect |
| replace Euclidean NMSE with CRE for bias analysis | Age-bias analysis, BFM and FLAME · Age vs CRE correlation (REALY) | Pearson r: strong age correlation with CRE (e.g., 0.77, 0.82 cheek) vs no significant bias under NMSE | Improved |
| pass 3DMM through 2D-to-3D reconstruction algorithms rather than direct 3D projection | 3DMM-based reconstruction · Age vs CRE correlation, nose | Pearson r | No clear effect |
Closest papers by meaning
- 3D Reconstruction of “In-the-Wild” Faces in Images and Videos (TPAMI 2018 · similarity 0.89)
- Inequality-Constrained and Robust 3D Face Model Fitting (ECCV 2020 · similarity 0.89)
- REALY: Rethinking the Evaluation of 3D Face Reconstruction (ECCV 2022 · similarity 0.89)
- Learning Complete 3D Morphable Face Models From Images and Videos (CVPR 2021 · similarity 0.89)
- What Does 2D Geometric Information Really Tell Us About 3D Face Shape? (IJCV 2019 · similarity 0.89)
- GIQ: Benchmarking 3D Geometric Reasoning of Vision Foundation Models with Simulated and Real Polyhedra (ICLR 2026 · 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.
