Research map / Detection and segmentation / CNN features
Semantic Alignment: Finding Semantically Consistent Ground-Truth for Facial Landmark Detection
Check 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 |
|---|---|---|---|
| Add Semantic Alignment to Hourglass training | Hourglass facial landmark detector · 300W Full | NME (%): 5.04 -> 4.37 | Helped |
| Add Semantic Alignment to Hourglass training | Hourglass facial landmark detector · AFLW-Full | NME (%): 1.95 -> 1.62 | Helped |
| Add Semantic Alignment to Hourglass training | Hourglass facial landmark detector · 300-VW Categories 1, 2, and 3 | RMSE: 4.32 -> 4.06; 3.83 -> 3.58; 9.91 -> 9.19 | Helped |
| Add GHCU after Semantic Alignment | Hourglass facial landmark detector · 300-VW Category 3 | RMSE: 9.19 -> 7.51 | Helped |
| Add GHCU after Semantic Alignment | Hourglass facial landmark detector · 300W Full | NME (%): 4.37 -> 4.35 | No clear effect |
| Add in-plane rotation normalization to Semantic Alignment and GHCU | Hourglass facial landmark detector · 300W Full | NME (%): 4.35 -> 4.02 | Helped |
| Remove Semantic Alignment by setting σ₂² / σ₁² to 0 (ablation) | Hourglass + Semantic Alignment · 300W Full | NME (%): 4.37 -> 4.99 | Hurt |
| Set σ₂² / σ₁² to 0.1 (ablation) | Hourglass + Semantic Alignment · 300W Full | NME (%): 4.99 -> 4.37 | Helped |
| Vary the Gaussian template size (ablation) | Hourglass + Semantic Alignment · 300W Full and AFLW-Full | NME (%): 300W: 4.76 (size=1), 4.72 (size=7), 4.61 (size=11), 4.53 (size=15), 4.37 (size=19), 4.43 (size=25); AFLW: 1.89 (size=1), | Mixed |
| Update the observation iteratively instead of keeping human annotations fixed (ablation) | Semantic Alignment training · 300W Full | NME | Helped |
| Train with labels searched by Semantic Alignment instead of human annotations (ablation) | Hourglass facial landmark detector · 300W Full | NME (%): 5.04 -> 4.37 | Helped |
| Train with labels searched by Semantic Alignment instead of human annotations (ablation) | ResNet coordinate regressor · 300W Full | NME (%): 5.49 -> 5.12 | Helped |
Closest papers by meaning
- LUVLi Face Alignment: Estimating Landmarks' Location, Uncertainty, and Visibility Likelihood (CVPR 2020 · similarity 0.89)
- Brute-Force Facial Landmark Analysis With a 140,000-Way Classifier (AAAI 2018 · similarity 0.89)
- Robust Facial Landmark Detection via a Fully-Convolutional Local-Global Context Network (CVPR 2018 · similarity 0.88)
- STAR Loss: Reducing Semantic Ambiguity in Facial Landmark Detection (CVPR 2023 · similarity 0.88)
- FaceLift: Semi-supervised 3D Facial Landmark Localization (CVPR 2024 · similarity 0.86)
- Semantic-Aware Transformation-Invariant RoI Align (AAAI 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.