Research map / Label-efficient and robust learning / Semi-supervised learning
Towards Semi-Supervised Deep Facial Expression Recognition With an Adaptive Confidence Margin
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use Ada-CM instead of FixMatch | ResNet-18 semi-supervised facial expression recognition · RAF-DB, 100 labels | accuracy: 60.67{ $\pm$2.25} -> 62.36{ $\pm$1.10} | Helped |
| use Ada-CM instead of FixMatch | ResNet-18 semi-supervised facial expression recognition · SFEW, 400 labels | accuracy: 50.37{ $\pm$0.45} -> 52.43{ $\pm$0.67} | Helped |
| use Ada-CM instead of FixMatch | ResNet-18 semi-supervised facial expression recognition · AffectNet, 10000 labels | accuracy: 56.50{ $\pm$0.43} -> 57.42{ $\pm$0.43} | Helped |
| use Ada-CM rather than MarginMix | WideResNet-28-2 semi-supervised facial expression recognition · RAF-DB, 400 labels | accuracy: 45.75 -> 59.03{ $\pm$0.73} | Helped |
| apply Ada-CM to cross-dataset training instead of FixMatch | ResNet-18 trained on RAF-DB and evaluated on CK+ · CK+, 4000 labels | accuracy: 84.18{ $\pm$0.99} -> 85.32{ $\pm$0.98} | Helped |
| add the contrastive objective for low-confidence unlabeled samples (ablation) | Ada-CM pseudo-label training · RAF-DB, 100 labels | accuracy: 61.50{ $\pm$2.10} -> 62.36{ $\pm$1.10} | Helped |
| add the contrastive objective for low-confidence unlabeled samples (ablation) | Ada-CM pseudo-label training · SFEW, 400 labels | accuracy: 51.04{ $\pm$0.58} -> 52.43{ $\pm$0.67} | Helped |
| add the pseudo-label loss for high-confidence unlabeled samples (ablation) | Ada-CM contrastive-only training · RAF-DB, 100 labels | accuracy: 54.27{ $\pm$2.79} -> 62.36{ $\pm$1.10} | Helped |
| add the pseudo-label loss for high-confidence unlabeled samples (ablation) | Ada-CM contrastive-only training · SFEW, 400 labels | accuracy: 45.99{ $\pm$0.35} -> 52.43{ $\pm$0.67} | Helped |
| adapt confidence thresholds by facial-expression category using correctly predicted labeled samples (ablation) | high-confidence pseudo-label selection · RAF-DB, 100 labels | accuracy: 60.67{ $\pm$2.25} -> 61.50{ $\pm$2.10} | Helped |
| adapt confidence thresholds by facial-expression category using correctly predicted labeled samples (ablation) | high-confidence pseudo-label selection · SFEW, 400 labels | accuracy: 50.37{ $\pm$0.45} -> 51.04{ $\pm$0.58} | Helped |
| vary the fixed confidence threshold from 0.5 to 0.8 to 0.95 (ablation) | FixMatch · RAF-DB, 100 labels | accuracy: 57.49{ $\pm$1.77} -> 58.94{ $\pm$2.05} -> 60.67{ $\pm$2.25} | Helped |
Closest papers by meaning
- Navigating Label Ambiguity for Facial Expression Recognition in the Wild (AAAI 2025 · similarity 0.88)
- CLEX: Complementary Label Exchange Learning for Noisy Facial Expression Recognition (CVPR 2026 · similarity 0.87)
- Relative Uncertainty Learning for Facial Expression Recognition (NeurIPS 2021 · similarity 0.87)
- D^3FER: Dual Channel and Dual Branch Network for Robust Facial Expression Recognition under Dual Challenges (CVPR 2026 · similarity 0.86)
- Dynamic Label Noise Suppression with Optimal Teacher Pool for Facial Expression Recognition (CVPR 2026 · similarity 0.86)
- Open-Set Facial Expression Recognition (AAAI 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.