Research map / Label-efficient and robust learning / Recognition and long-tail
PIAP-DF: Pixel-Interested and Anti Person-Specific Facial Action Unit Detection Net With Discrete Feedback Learning
Yang Tang, Wangding Zeng, Dafei Zhao, Honggang Zhang
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 stage-2 Pixel-Interested learning (learned PI maps) (ablation) | EfficientNet-B1 baseline · BP4D AU detection | average F1: 57.7 -> 60.5 | Improved |
| add stage-3 PI decoder supervision on top of stage 2 (ablation) | PI-2 · BP4D AU detection | average F1: 60.5 -> 63.9 | Improved |
| add stage-2 PI learning (ablation) | EfficientNet-B1 baseline · BP4D AU2 | F1: 42.8 -> 21.8 | Hurt |
| add Anti Person-Specific learning (ablation) | EfficientNet-B1 baseline · BP4D AU detection | average F1: 57.7 -> 61.9 | Improved |
| add Discrete Feedback semi-supervised learning (ablation) | EfficientNet-B1 baseline · BP4D AU detection | average F1: +2.8 | Improved |
| add Anti Person-Specific learning to PI-3 (ablation) | PI-3 · BP4D AU detection | average F1: 63.9 -> 64.1 | Improved |
| add Discrete Feedback to PIAP (ablation) | PIAP · BP4D AU detection | average F1: 64.1 -> 64.4 | Improved |
| orthogonalize AU features to face-recognition features instead of landmark features | Anti Person-Specific learning · BP4D AU detection | average F1 | Improved |
| use PIAP-DF instead of best prior J^AA-Net | BP4D AU detection · BP4D AU detection | average F1: 62.4 -> 64.4 | Improved |
| use PIAP (fine-tuned from BP4D) instead of best prior J^AA-Net | DISFA AU detection · DISFA AU detection | average F1: 63.5 -> 63.8 | Improved |
Closest papers by meaning
- Local Relationship Learning With Person-Specific Shape Regularization for Facial Action Unit Detection (CVPR 2019 · similarity 0.87)
- Deep Region and Multi-Label Learning for Facial Action Unit Detection (CVPR 2016 · similarity 0.86)
- Action Unit Detection With Region Adaptation, Multi-Labeling Learning and Optimal Temporal Fusing (CVPR 2017 · similarity 0.86)
- Facial Action Unit Detection With Transformers (CVPR 2021 · similarity 0.86)
- Facial Action Unit Detection by Adaptively Constraining Self-Attention and Causally Deconfounding Sample (IJCV 2024 · similarity 0.86)
- AUFormer: Vision Transformers are Parameter-Efficient Facial Action Unit Detectors (ECCV 2024 · 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.
