Research map / Label-efficient and robust learning / Semi-supervised learning
Laplace Landmark Localization
Joseph P. Robinson, Yuncheng Li, Ning Zhang, Yun Fu, Sergey Tulyakov
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 softargmax objective with LaplaceKL (ablation) | RCN landmark detector · 300W full | NMSE: 4.25 -> 4.01 | Improved |
| replace softargmax with LaplaceKL on AFLW (ablation) | RCN landmark detector · AFLW | NMSE: 2.26 -> 1.97 | Improved |
| add adversarial training with 70K unlabeled MegaFace images (ablation) | LaplaceKL · 300W full | NMSE: 4.01 -> 3.91 | Improved |
| add adversarial training with 70K unlabeled images (ablation) | Softargmax · 300W full | NMSE: 4.25 -> 4.29 | No clear effect |
| add adversarial training with 70K unlabeled images (ablation) | LaplaceKL · 300W challenge | NMSE: 7.01 -> 6.87 | Improved |
| use 300W itself as unlabeled set in adversarial branch (ablation) | LaplaceKL · 300W full | NMSE: 4.01 -> 4.06 | No clear effect |
| use 300W itself as unlabeled set in adversarial branch (ablation) | Softargmax · 300W full | NMSE: 4.24 -> 4.26 | No clear effect |
| reduce channels to 1/8 (0.0389M params) (ablation) | Softargmax · 300W full | NMSE: 4.25 -> 6.86 | Hurt |
| reduce channels to 1/8 (0.0389M params) (ablation) | LaplaceKL+D(70K) · 300W full | NMSE: 3.91 -> 4.85 | Hurt |
| reduce channels to 1/16 (0.0174M params) (ablation) | LaplaceKL · 300W full | NMSE: 4.01 -> 7.38 | Hurt |
| LaplaceKL+D(70K) vs Softargmax+D(70K) at 1/16 channels (ablation) | small RCN · 300W full (1/16 channels) | NMSE: 9.02 -> 7.01 | Improved |
| add 70K unlabeled adversarial data to Softargmax at 1/16 channels (ablation) | Softargmax · 300W full (1/16 channels) | NMSE: 9.79 -> 9.02 | Improved |
Closest papers by meaning
- Improving Landmark Localization With Semi-Supervised Learning (CVPR 2018 · similarity 0.87)
- Multi-Domain Multi-Definition Landmark Localization for Small Datasets (ECCV 2022 · similarity 0.87)
- Wing Loss for Robust Facial Landmark Localisation With Convolutional Neural Networks (CVPR 2018 · similarity 0.86)
- Heatmap Regression via Randomized Rounding (TPAMI 2021 · similarity 0.85)
- FaceLift: Semi-supervised 3D Facial Landmark Localization (CVPR 2024 · similarity 0.85)
- AnchorFace: An Anchor-based Facial Landmark Detector Across Large Poses (AAAI 2021 · 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.
