Research map / Learning theory and algorithms / Prediction and decision losses
Adaptive Wing Loss for Robust Face Alignment via Heatmap Regression
Xinyao Wang, Liefeng Bo, Li Fuxin
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 |
|---|---|---|---|
| replace MSE loss with Adaptive Wing loss (ablation) | MSE baseline · WFLW (GT-cropped) | NME (%): 5.39 -> 4.65 | Improved |
| add Weighted Loss Map to AWing (ablation) | AW · WFLW (GT-cropped) | NME (%): 4.65 -> 4.30 | Improved |
| add Weighted Loss Map to MSE (ablation) | MSE · WFLW (GT-cropped) | NME (%): 5.39 -> 5.04 | Improved |
| use Weighted Loss Map instead of naive weight map (ablation) | AW+WM · WFLW (GT-cropped) | NME (%): 4.49 -> 4.30 | Improved |
| add boundary prediction (ablation) | AW+WM · WFLW (GT-cropped) | NME (%): 4.30 -> 4.28 | Improved |
| add CoordConv (ablation) | AW+WM+B · WFLW (GT-cropped) | NME (%): 4.28 -> 4.26 | Improved |
| add CoordConv with boundary coordinates (ablation) | AW+WM+B+C · WFLW (GT-cropped) | NME (%): 4.26 -> 4.21 | Improved |
| vary omega and epsilon in AWing (ablation) | AWing · WFLW (GT-cropped) | NME (%): 4.21 to 4.30 | No clear effect |
| vary theta from 0.3 to 0.7 (ablation) | AWing · WFLW (GT-cropped) | NME (%): 4.25, 4.22, 4.21, 4.26, 4.23 | No clear effect |
| increase hourglass stacks from 1 to 4 (ablation) | AWing · WFLW (GT-cropped) | NME (%): 4.50 -> 4.21 | Improved |
| train AWing on LSP instead of MSE (ablation) | 4HG · LSP | PCK@0.2: 81.8 -> 85.9 | Improved |
| add label noise (30% points, 30% shift) to WFLW (ablation) | AWing · WFLW (noisy labels) | NME (%): 4.65 -> 4.86 | Hurt |
Closest papers by meaning
- PropagationNet: Propagate Points to Curve to Learn Structure Information (CVPR 2020 · similarity 0.91)
- Rectified Wing Loss for Efficient and Robust Facial Landmark Localisation with Convolutional Neural Networks (IJCV 2019 · similarity 0.91)
- Wing Loss for Robust Facial Landmark Localisation With Convolutional Neural Networks (CVPR 2018 · similarity 0.90)
- Robust Face Alignment via Deep Progressive Reinitialization and Adaptive Error-driven Learning (TPAMI 2021 · similarity 0.89)
- Heatmap Regression without Soft-Argmax for Facial Landmark Detection (ICCV 2025 · similarity 0.88)
- POPoS: Improving Efficient and Robust Facial Landmark Detection with Parallel Optimal Position Search (AAAI 2025 · similarity 0.86)
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.
