Research map / 3D vision / Face and shape
Robust Model-Based Face Reconstruction Through Weakly-Supervised Outlier Segmentation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| jointly train face reconstruction with weakly supervised outlier segmentation | model-based face autoencoder · NoW testset | median reconstruction error (mm): 1.23 -> 1.04 | Helped |
| jointly train face reconstruction with weakly supervised outlier segmentation | model-based face autoencoder · CelebA-Occluded | RMSE: 8.79 ± 0.45 -> 8.71 ± 0.48 | Helped |
| add the in-domain misfit prior (ablation) | FOCUS · NoW testset | mean reconstruction error (mm): 1.30 -> 1.28 | Helped |
| add the in-domain misfit prior (ablation) | FOCUS · NoW evaluation set | mean reconstruction error (mm): 1.31 -> 1.29 | Helped |
| include the segmentation network in the reconstruction pipeline (ablation) | pretrained face autoencoder · NoW evaluation set | median reconstruction error (mm): 1.06 -> 1.05 | Helped |
| include the segmentation network in the reconstruction pipeline (ablation) | pretrained face autoencoder · AR-scarf | F1: 0.73 ± 0.07 -> 0.88 ± 0.05 | Helped |
| add neighbor and perceptual losses to segmentation training (ablation) | FOCUS segmentation network · AR-scarf | F1: 0.82 ± 0.07 -> 0.88 ± 0.05 | Helped |
| use only the neighbor loss (ablation) | FOCUS segmentation network · AR-scarf | F1: 0.82 ± 0.07 -> 0.85 ± 0.06 | Helped |
| use only the perceptual losses (ablation) | FOCUS segmentation network · AR-scarf | F1: 0.82 ± 0.07 -> 0.90 ± 0.05 | Helped |
| replace neighbor and perceptual losses with only the neighbor loss (ablation) | FOCUS segmentation network · AR-scarf | F1: 0.90 ± 0.05 -> 0.85 ± 0.06 | Hurt |
| replace neighbor and perceptual losses with only the perceptual losses (ablation) | FOCUS segmentation network · AR-scarf | F1: 0.85 ± 0.06 -> 0.90 ± 0.05 | Helped |
Closest papers by meaning
- A Robust Multilinear Model Learning Framework for 3D Faces (CVPR 2016 · similarity 0.87)
- Self-Supervised Multi-Level Face Model Learning for Monocular Reconstruction at Over 250 Hz (CVPR 2018 · similarity 0.87)
- Extreme 3D Face Reconstruction: Seeing Through Occlusions (CVPR 2018 · similarity 0.87)
- OFER: Occluded Face Expression Reconstruction (CVPR 2025 · similarity 0.87)
- SegFace: Face Segmentation of Long-Tail Classes (AAAI 2025 · similarity 0.85)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · 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.