Research map / Label-efficient and robust learning / Recognition and long-tail
Deep Structure Inference Network for Facial Action Unit Recognition
Ciprian Corneanu, Meysam Madadi, Sergio Escalera
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 class-balanced weighted loss to patch prediction (ablation) | PP(face) · BP4D | F1-frame avg: 55.3 -> 56.9 | Improved |
| use proposed shallow CNN instead of fine-tuned VGG-16 face model (ablation) | patch prediction · BP4D | F1-frame avg: 46.5 -> 56.9 | Improved |
| add per-AU patch fusion on top of whole-face patch prediction (ablation) | PP(face) · BP4D | F1-frame avg: 56.9 -> 57.8 | Improved |
| add per-AU patch fusion on top of whole-face patch prediction (ablation) | PP(face) · DISFA | F1-frame avg: 43.9 -> 48.0 | Improved |
| add structure inference (message passing, T=10) after fusion (ablation) | PP+F · DISFA | F1-frame avg: 48.0 -> 53.6 | Improved |
| add structure inference (message passing, T=10) after fusion (ablation) | PP+F · BP4D | F1-frame avg: 57.8 -> 58.9 | Improved |
| increase structure inference iterations from 2 to 10 (ablation) | DSIN · BP4D | F1-frame avg: 58.2 -> 58.5 -> 58.9 | Improved |
| add correction factors (mutual gating) to message passing (ablation) | DSIN T=2 · BP4D | F1-frame avg: 57.9 -> 58.2 | Improved |
| tune decision threshold per AU on validation set | DSIN · BP4D | F1-frame avg: 58.9 -> 61.7 | Improved |
| tune decision threshold per AU on validation set | DSIN · DISFA | F1-frame avg: 53.6 -> 56.7 | Improved |
| add L1-style regularization on correction factors (r up to 1e-2) (ablation) | DSIN · BP4D | F1-frame | No clear effect |
| raise correction-factor regularization above 5e-2 (ablation) | DSIN · BP4D | F1-frame | Hurt |
Closest papers by meaning
- Knowledge Augmented Deep Neural Networks for Joint Facial Expression and Action Unit Recognition (NeurIPS 2020 · similarity 0.93)
- Semantic Relationships Guided Representation Learning for Facial Action Unit Recognition (AAAI 2019 · similarity 0.91)
- Deep Facial Action Unit Recognition From Partially Labeled Data (ICCV 2017 · similarity 0.91)
- Facial Action Unit Detection With Transformers (CVPR 2021 · similarity 0.90)
- Deep Structured Learning for Facial Action Unit Intensity Estimation (CVPR 2017 · similarity 0.90)
- Weakly Supervised Facial Action Unit Recognition Through Adversarial Training (CVPR 2018 · similarity 0.90)
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.
