Research map / 3D vision / Face and shape
How Far Are We From Solving the 2D & 3D Face Alignment Problem? (And a Dataset of 230,000 3D Facial Landmarks)
Adrian Bulat, Georgios Tzimiropoulos
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 |
|---|---|---|---|
| increase 3D-FAN size from 2M to 24M parameters (ablation) | 3D-FAN · LS3D-W Balanced (yaw 60-90) | AUC@7%: 55.8% -> 68.8% | Improved |
| increase 3D-FAN size from 2M to 24M parameters (ablation) | 3D-FAN · LS3D-W Balanced (yaw 0-30) | AUC@7%: 70.9% -> 73.5% | Improved |
| increase 3D-FAN size from 6M to 12M parameters (ablation) | 3D-FAN · LS3D-W Balanced (yaw 60-90) | AUC@7%: 58.3% -> 67.1% | Improved |
| increase 3D-FAN size from 12M to 24M parameters (ablation) | 3D-FAN · LS3D-W Balanced (yaw 0-30) | AUC@7%: 72.7% -> 73.5% | No clear effect |
| raise initialization noise from 0% to 30% (ablation) | 3D-FAN · LS3D-W Balanced (yaw 0-30) | AUC@7%: 74.5% -> 63.8% | Hurt |
| raise initialization noise from 0% to 20% (ablation) | 3D-FAN · LS3D-W Balanced (yaw 60-90) | AUC@7%: 69.8% -> 66.1% | Hurt |
| move from yaw 0-30 to yaw 60-90 faces | 3D-FAN · LS3D-W Balanced | AUC@7%: 73.5% -> 68.8% | Hurt |
| reduce face resolution to 30 pixels (ablation) | 3D-FAN · LS3D-W Balanced | AUC@7% | Hurt |
| use 2D-FAN instead of MDM | 2D alignment · 300VW-C | AUC@7%: 54.6% -> 64.1% | Improved |
| use 2D-FAN instead of MDM | 2D alignment · 300-W test | AUC@7%: 58.1% -> 66.9% | Improved |
| use 2D-FAN instead of MDM | 2D alignment · Menpo | AUC@7%: 67.1% -> 67.5% | Improved |
Closest papers by meaning
- Faster Than Real-Time Facial Alignment: A 3D Spatial Transformer Network Approach in Unconstrained Poses (ICCV 2017 · similarity 0.93)
- Large Pose 3D Face Reconstruction From a Single Image via Direct Volumetric CNN Regression (ICCV 2017 · similarity 0.92)
- A Simple, Fast and Highly-Accurate Algorithm to Recover 3D Shape from 2D Landmarks on a Single Image (TPAMI 2017 · similarity 0.91)
- Face Alignment Across Large Poses: A 3D Solution (CVPR 2016 · similarity 0.91)
- LUVLi Face Alignment: Estimating Landmarks' Location, Uncertainty, and Visibility Likelihood (CVPR 2020 · similarity 0.91)
- FaceLift: Semi-supervised 3D Facial Landmark Localization (CVPR 2024 · similarity 0.90)
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.
