Research map / 3D vision / Face and shape
Self-Supervised 3D Face Reconstruction via Conditional Estimation
Yandong Wen, Weiyang Liu, Bhiksha Raj, Rita Singh
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 |
|---|---|---|---|
| estimate shape, reflectance and illumination conditionally on previously derived parameters instead of independently (ablation) | CEST · CelebA (1,000 images) | photometric error: 13.76 -> 10.74 | Improved |
| add reflectance symmetry constraint (ablation) | CEST · VoxCeleb1 qualitative | reflectance and illumination separation | Improved |
| add reflectance consistency across video clip (ablation) | CEST · VoxCeleb1 qualitative | reflectance and illumination decomposition | Improved |
| add both symmetry and consistency constraints versus none (ablation) | CEST · VoxCeleb1 qualitative | disentanglement of parameters | Improved |
| use averaged video-clip reflectance with 2, 4 or 8 frames per clip instead of per-frame cross reconstruction (ablation) | CEST · VoxCeleb1 qualitative | reflectance disentanglement | Hurt |
| apply reflectance consistency across whole video instead of short clip (ablation) | CEST · VoxCeleb1 qualitative | reflectance quality | No clear effect |
| train CEST self-supervised on videos versus fully supervised 3DDFA | CEST · AFLW2000-3D | NME: 5.37 -> 3.37 | Improved |
| CEST versus N3DMM with dense-landmark pretraining | CEST · MICC Florence | NME: 3.20 -> 3.14 | Improved |
Closest papers by meaning
- Learning Complete 3D Morphable Face Models From Images and Videos (CVPR 2021 · similarity 0.89)
- Self-Supervised Multi-Level Face Model Learning for Monocular Reconstruction at Over 250 Hz (CVPR 2018 · similarity 0.88)
- FML: Face Model Learning From Videos (CVPR 2019 · similarity 0.86)
- Self-Supervised Monocular 3D Face Reconstruction by Occlusion-Aware Multi-view Geometry Consistency (ECCV 2020 · similarity 0.86)
- Face Anything: 4D Face Reconstruction from Any Image Sequence (ECCV 2026 · similarity 0.86)
- 3D Facial Expressions through Analysis-by-Neural-Synthesis (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.
