Research map / Label-efficient and robust learning / Recognition and long-tail
ADNet: Leveraging Error-Bias Towards Normal Direction in Face Alignment
Yangyu Huang, Hao Yang, Chong Li, Jongyoo Kim, Fangyun Wei
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 |
|---|---|---|---|
| add ADL with lambda=2.0 (ablation) | baseline · 300W fullset (no AAM) | NME: 3.38 -> 3.23 | Improved |
| weight tangent direction more (lambda=0.5) instead of normal (ablation) | ADL · 300W fullset (no AAM) | NME: 3.38 -> 3.43 | Hurt |
| raise lambda from 2.0 to 8.0 (ablation) | ADL · 300W fullset (no AAM) | NME: 3.23 -> 3.27 | No clear effect |
| add AAM (anisotropic attention) (ablation) | baseline · 300W fullset (no ADL) | NME: 3.38 -> 2.98 | Improved |
| use AAM instead of point attention (ablation) | attention module · 300W fullset (no ADL) | NME: 3.12 -> 2.98 | Improved |
| use AAM instead of edge attention (ablation) | attention module · 300W fullset (no ADL) | NME: 3.09 -> 2.98 | Improved |
| add ADL and AAM (ablation) | Stacked 4 HGNet · 300W fullset | NME: 3.38 -> 2.93 | Improved |
| add ADL and AAM (ablation) | Single HGNet · 300W fullset | NME: 3.39 -> 2.99 | Improved |
| add ADL and AAM (ablation) | ResNet50 · 300W fullset | NME: 3.43 -> 3.11 | Improved |
| add ADL and AAM (ablation) | baseline · 300W normal-direction NME | NME: 1.91 -> 1.54 | Improved |
| add ADL and AAM (ablation) | baseline · 300W tangent-direction NME | NME: 2.55 -> 2.28 | Improved |
| add ADL and AAM, raising error bias rate (ablation) | baseline · 300W bias rate | bias rate: 33.51% -> 48.05% | Improved |
Later papers that compared with this method
6 later accepted papers report ADNet as a baseline in their results tables; 5 of them report a better number on most of the results they share with it.
- Semi-Supervised High Dynamic Range Image Reconstructing via Bi-Level Uncertain Area Masking (AAAI 2026 · better on 2 of 2 shared results)
- DARC: Dual Adjustment Reasoning with Counterfactuals for Trustworthy Chest X-ray Classification (CVPR 2026 · better on 1 of 1 shared result)
- Faceptor: A Generalist Model for Face Perception (ECCV 2024 · better on 1 of 1 shared result)
- FreeEnricher: Enriching Face Landmarks without Additional Cost (AAAI 2023 · better on 2 of 2 shared results)
- Multi-Domain Multi-Definition Landmark Localization for Small Datasets (ECCV 2022 · better on 6 of 7 shared results)
- Occlusion-Robust Face Alignment Using a Viewpoint-Invariant Hierarchical Network Architecture (CVPR 2022 · better on 1 of 5 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Facial Landmark Detection with Tweaked Convolutional Neural Networks (TPAMI 2017 · similarity 0.89)
- Robust Face Alignment via Deep Progressive Reinitialization and Adaptive Error-driven Learning (TPAMI 2021 · similarity 0.89)
- PropagationNet: Propagate Points to Curve to Learn Structure Information (CVPR 2020 · similarity 0.88)
- How Far Are We From Solving the 2D & 3D Face Alignment Problem? (And a Dataset of 230,000 3D Facial Landmarks) (ICCV 2017 · similarity 0.88)
- LUVLi Face Alignment: Estimating Landmarks' Location, Uncertainty, and Visibility Likelihood (CVPR 2020 · similarity 0.87)
- Semantic Alignment: Finding Semantically Consistent Ground-Truth for Facial Landmark Detection (CVPR 2019 · similarity 0.87)
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.
