Research map / Label-efficient and robust learning / Recognition and long-tail
Wing Loss for Robust Facial Landmark Localisation With Convolutional Neural Networks
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 |
|---|---|---|---|
| Replace L2 with L1 loss (ablation) | CNN-6 · AFLW-Full | average normalised error: 2.41$\times10^{-2}$ -> 2.00$\times10^{-2}$ | Helped |
| Replace L2 with smooth L1 loss (ablation) | CNN-6 · AFLW-Full | average normalised error: 2.41$\times10^{-2}$ -> 2.02$\times10^{-2}$ | Helped |
| Use Wing loss instead of L1 or smooth L1 (ablation) | CNN-6 · AFLW-Full | average normalised error: 2.00$\times10^{-2}$ -> 1.88$\times10^{-2}$ | Helped |
| Add pose-based data balancing (ablation) | CNN-6 with Wing loss · AFLW-Full | average normalised error ($\times 10^{-2}$): 1.88 -> 1.83 | Helped |
| Add the CNN-7 refinement stage (ablation) | CNN-6 + PDB with Wing loss · AFLW-Full | average normalised error ($\times 10^{-2}$): 1.83 -> 1.65 | Helped |
| Add pose-based data balancing (ablation) | CNN-6/7 with Wing loss · AFLW-Full | average normalised error ($\times 10^{-2}$): 1.71 -> 1.65 | Helped |
| Replace L2 with Wing loss (ablation) | CNN-6/7 + PDB · 300W Common | NME: 4.18 -> 3.27 | Helped |
| Replace L2 with Wing loss (ablation) | CNN-6/7 + PDB · 300W Challenge | NME: 8.19 -> 7.18 | Mixed |
| Replace L1 with Wing loss (ablation) | CNN-6/7 + PDB · 300W Challenge | NME: 7.02 -> 7.18 | Hurt |
| Replace CNN-6/7 with pretrained ResNet-50 (ablation) | PDB with Wing loss · AFLW-Full | NME ($\times10^{-2}$): 1.65 -> 1.47 | Helped |
| Replace Wing with L2 loss (ablation) | ResNet-50 · AFLW-Full | NME ($\times10^{-2}$): 1.47 -> 1.68 | Hurt |
| Vary Wing parameters w and epsilon (ablation) | CNN-6 · AFLW-Full | average normalised error ($\times 10^{-2}$): 1.88 to 2.02 | Mixed |
Closest papers by meaning
- AnchorFace: An Anchor-based Facial Landmark Detector Across Large Poses (AAAI 2021 · similarity 0.87)
- Disentangling 3D Pose in a Dendritic CNN for Unconstrained 2D Face Alignment (CVPR 2018 · similarity 0.86)
- LUVLi Face Alignment: Estimating Landmarks' Location, Uncertainty, and Visibility Likelihood (CVPR 2020 · similarity 0.86)
- Brute-Force Facial Landmark Analysis With a 140,000-Way Classifier (AAAI 2018 · similarity 0.86)
- FaceLift: Semi-supervised 3D Facial Landmark Localization (CVPR 2024 · similarity 0.84)
- POPoS: Improving Efficient and Robust Facial Landmark Detection with Parallel Optimal Position Search (AAAI 2025 · similarity 0.83)
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.