Research map / Detection and segmentation / CNN features
JÂA-Net: Joint Facial Action Unit Detection and Face Alignment Via Adaptive Attention
Zhiwen Shao, Zhilei Liu, Jianfei Cai, Lizhuang Ma
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 hierarchical and multi-scale region layer instead of region layer (ablation) | R-Net · BP4D (12 AUs) | average F1-frame: 54.9 -> 55.8 | Improved |
| add Dice coefficient loss (ablation) | H-Net · BP4D (12 AUs) | average F1-frame: 55.8 -> 56.6 | Improved |
| weight the loss of each AU (ablation) | HD-Net · BP4D (12 AUs) | average F1-frame: 56.6 -> 57.4 | Improved |
| add face alignment module (no feature integration) (ablation) | HDW-Net · BP4D (12 AUs) | average F1-frame: 57.4 -> 58.0 | Improved |
| integrate face alignment feature with global feature for AU detection (ablation) | J-Net w/o IF · BP4D (12 AUs) | average F1-frame: 58.0 -> 58.8 | Improved |
| add predefined attention maps for local AU features (ablation) | J-Net · BP4D (12 AUs) | average F1-frame: 58.8 -> 59.3 | Improved |
| add global feature (ablation) | JA-Net w/o GF · BP4D (12 AUs) | average F1-frame: 59.3 -> 59.9 | Improved |
| add AU attention refinement (ablation) | JA-Net · BP4D (12 AUs) | average F1-frame: 59.9 -> 61.0 | Improved |
| add local AU detection loss E_local_au (ablation) | JAA-Net w/o E_local_au · BP4D (12 AUs) | average F1-frame: 61.0 -> 62.4 | Improved |
| add back-propagation enhancement (ablation) | JAA-Net w/o E_local_au · BP4D (12 AUs) | average F1-frame: 61.0 -> 61.6 | Improved |
| add refinement constraint E_r on top of back-propagation enhancement (ablation) | JAA-Net w/o E_local_au+BE · BP4D (12 AUs) | average F1-frame: 61.6 -> 60.8 | Hurt |
| increase enhancement coefficient lambda_e from 2 to 3 (ablation) | JAA-Net w/o E_local_au+BE · BP4D (12 AUs) | average F1-frame: 61.6 -> 61.1 | Hurt |
Closest papers by meaning
- Deep Adaptive Attention for Joint Facial Action Unit Detection and Face Alignment (ECCV 2018 · similarity 0.97)
- Facial Action Unit Detection by Adaptively Constraining Self-Attention and Causally Deconfounding Sample (IJCV 2024 · similarity 0.89)
- EAC-Net: Deep Nets with Enhancing and Cropping for Facial Action Unit Detection (TPAMI 2018 · similarity 0.89)
- Action Unit Detection With Region Adaptation, Multi-Labeling Learning and Optimal Temporal Fusing (CVPR 2017 · similarity 0.88)
- Trend-Aware Supervision: On Learning Invariance for Semi-supervised Facial Action Unit Intensity Estimation (AAAI 2024 · similarity 0.86)
- Facial-R1: Aligning Reasoning and Recognition for Facial Emotion Analysis (AAAI 2026 · 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.
