Research map / Label-efficient and robust learning / Recognition and long-tail
Suppressing Uncertainties for Large-Scale Facial Expression Recognition
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 |
|---|---|---|---|
| add SCN uncertainty suppression | ResNet-18 on RAF-DB with 30% synthetic label noise, trained from scratch · RAF-DB, 30% synthetic label noise | accuracy: 46.81 -> 60.61; 13.80% | Helped |
| add SCN uncertainty suppression | ResNet-18 on FERPlus with 30% synthetic label noise, trained from scratch · FERPlus, 30% synthetic label noise | accuracy: 68.54 -> 70.45; 1.91% | Helped |
| add SCN uncertainty suppression | ResNet-18 on AffectNet with 30% synthetic label noise, trained from scratch · AffectNet, 30% synthetic label noise | accuracy: 38.35 -> 39.42; 1.07% | Helped |
| add SCN uncertainty suppression | pretrained ResNet-18 on RAF-DB with 30% synthetic label noise · RAF-DB, 30% synthetic label noise, pretrained | accuracy: 75.26 -> 77.46; 2.20% | Helped |
| add SCN uncertainty suppression | pretrained ResNet-18 on AffectNet with 30% synthetic label noise · AffectNet, 30% synthetic label noise, pretrained | accuracy: 52.58 -> 55.05; 2.47% | Helped |
| add SCN uncertainty suppression | pretrained ResNet-18 on FERPlus with 30% synthetic label noise · FERPlus, 30% synthetic label noise, pretrained | accuracy: 79.34 -> 82.47; 3.12% | Helped |
| add self-attention importance weighting (ablation) | ResNet-18 baseline on RAF-DB · RAF-DB | accuracy: 72.00 -> 76.26 | Helped |
| add rank regularization to self-attention weighting (ablation) | SCN on RAF-DB · RAF-DB | accuracy: 76.26 -> 76.57 | Helped |
| add careful relabeling to weighting and rank regularization (ablation) | SCN on RAF-DB · RAF-DB | accuracy: 76.57 -> 78.31 | Helped |
| add naive relabeling without sample weighting or ranking (ablation) | ResNet-18 baseline on RAF-DB · RAF-DB | accuracy: 72.00 -> 71.25 | Hurt |
| increase RR-Loss weight from gamma 0.5 to 0.8 (ablation) | SCN on RAF-DB · RAF-DB | accuracy: 78.31% -> 71.75% | Hurt |
| pretrain with SCN on noisy WebEmotion rather than without SCN | model fine-tuned on RAF-DB · RAF-DB after WebEmotion pretraining | accuracy: 80.42% -> 82.45% | Helped |
Closest papers by meaning
- Robust Lightweight Facial Expression Recognition Network with Label Distribution Training (AAAI 2021 · similarity 0.87)
- FAIR-FER: A Latent Alignment Approach for Mitigating Bias in Facial Expression Recognition (Student Abstract) (AAAI 2024 · similarity 0.85)
- Navigating Label Ambiguity for Facial Expression Recognition in the Wild (AAAI 2025 · similarity 0.85)
- Open-Set Facial Expression Recognition (AAAI 2024 · similarity 0.85)
- Hypergraph-Guided Disentangled Spectrum Transformer Networks for Near-Infrared Facial Expression Recognition (AAAI 2024 · similarity 0.84)
- D2SP: Dynamic Dual-Stage Purification Framework for Dual Noise Mitigation in Vision-based Affective Recognition. (CVPR 2025 · 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.