Research map / Label-efficient and robust learning / Recognition and long-tail
Peak-Piloted Deep Network for Facial Expression Recognition
Xiangyun Zhao, Xiaodan Liang, Luoqi Liu, Teng Li, Yugang Han, Nuno Vasconcelos, Shuicheng Yan
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 peak-piloted feature supervision with PGS to GoogLeNet | GoogLeNet · CK+ (6 expressions, last three frames) | accuracy: 95.0% -> 97.3% | Improved |
| add peak-piloted feature supervision with PGS to GoogLeNet | GoogLeNet · Oulu-CASIA (UDCS setting) | accuracy: 66.6% -> 72.4% | Improved |
| add peak gradient suppression instead of standard SGD (ablation) | PPDN (standard SGD) · CK+ (weak expression test set) | accuracy: 81.34% -> 83.36% | Improved |
| add peak gradient suppression instead of standard SGD (ablation) | PPDN (standard SGD) · CK+ (combined test set) | accuracy: 94.18% -> 95.33% | Improved |
| add peak gradient suppression instead of standard SGD (ablation) | PPDN (standard SGD) · Oulu-CASIA (peak expression test set) | accuracy: 82.91% -> 84.59% | Improved |
| add peak gradient suppression instead of standard SGD (ablation) | PPDN (standard SGD) · Oulu-CASIA (weak expression test set) | accuracy: 67.05% -> 67.95% | Improved |
| add peak-piloted training to GoogLeNet | GoogLeNet · CK+ (weak expression test set) | accuracy: 78.10% -> 83.36% | Improved |
| add peak-piloted training to GoogLeNet | GoogLeNet · Oulu-CASIA (weak expression test set) | accuracy: 64.64% -> 67.95% | Improved |
| use ground-truth peak frame at test time instead of highest-score frame (ablation) | PPDN w/o peak · Oulu-CASIA (sequence-level) | accuracy: 83.67% -> 84.59% | Improved |
| use ground-truth peak frame at test time instead of highest-score frame (ablation) | PPDN w/o peak · CK+ (sequence-level) | accuracy: 99.2% -> 99.3% | No clear effect |
| apply peak-piloted loss to both FC layers instead of only the last FC layer (ablation) | PPDN · Oulu-CASIA (peak-piloted layer ablation) | accuracy: 73.33% -> 74.99% | Improved |
| apply peak-piloted loss to both FC layers instead of only the first FC layer (ablation) | PPDN · Oulu-CASIA (peak-piloted layer ablation) | accuracy: 73.48% -> 74.99% | Improved |
Closest papers by meaning
- Intensity-Aware Loss for Dynamic Facial Expression Recognition in the Wild (AAAI 2023 · similarity 0.87)
- Deep Structured Learning for Facial Action Unit Intensity Estimation (CVPR 2017 · similarity 0.87)
- Suppressing Uncertainties for Large-Scale Facial Expression Recognition (CVPR 2020 · similarity 0.87)
- Deep Multi-Task Learning to Recognise Subtle Facial Expressions of Mental States (ECCV 2018 · similarity 0.86)
- D^3FER: Dual Channel and Dual Branch Network for Robust Facial Expression Recognition under Dual Challenges (CVPR 2026 · similarity 0.86)
- Hypergraph-Guided Disentangled Spectrum Transformer Networks for Near-Infrared Facial Expression Recognition (AAAI 2024 · similarity 0.84)
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.
