Research map / Label-efficient and robust learning / Noisy labels
Teaching with Soft Label Smoothing for Mitigating Noisy Labels in Facial Expressions
Tohar Lukov, Na Zhao, Gim Hee Lee, Ser-Nam Lim
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 SLS over teacher logits to mean-teacher (ablation) | MT+U · RAF-DB 30% symmetric noise | accuracy: 85.13 -> 86.94 | Improved |
| add mean-teacher to SLS (ablation) | SLS vanilla · RAF-DB 30% symmetric noise | accuracy: 85.33 -> 86.82 | Improved |
| add uncertainty module to MT+SLS (ablation) | MT+SLS · RAF-DB 30% symmetric noise | accuracy: 86.82 -> 86.94 | Improved |
| use instance-aware SLS instead of LSR* (ablation) | teacher smoothing · RAF-DB 30% symmetric noise | accuracy: 85.85 -> 86.94 | Improved |
| use non-zero low confidences (epsilon>0) in SLS (ablation) | SLS · RAF-DB 30% symmetric noise | accuracy: 85.30 -> 86.94 | Improved |
| apply SLS to teacher logits instead of student logits (ablation) | SLS · RAF-DB 30% symmetric noise | accuracy: 85.23 -> 86.94 | Improved |
| apply SLS to both student and teacher logits instead of teacher only (ablation) | SLS · RAF-DB 30% symmetric noise | accuracy: 86.94 -> 85.36 | Hurt |
| add SLS to MT in noisy-part analysis (ablation) | MT+U · RAF-DB 30% symmetric noise | noisy samples predicted with injected wrong label: 28.33 -> 13.57 | Improved |
| increase epsilon beyond 0.3 (ablation) | SLS · RAF-DB 30% symmetric noise | accuracy | Hurt |
| raise tau toward 0.4 so only one high-confidence class remains (ablation) | SLS · RAF-DB 30% symmetric noise | accuracy | Hurt |
| use SOFT instead of DMUE under symmetric noise | SOFT · RAF-DB symmetric noise 10-70% | accuracy: 5.7%, 7%, 7.5% and 26.5% on RAF-DB with 10%, 20%, 30% and 70% noise rates | Improved |
| use SOFT instead of DMUE under symmetric noise | SOFT · AffectNet-8 symmetric noise | accuracy: 0.1%, 1.9%, 2.8% on AffectNet-8 with 10%, 20%, 30% noise rates | Improved |
Closest papers by meaning
- Navigating Label Ambiguity for Facial Expression Recognition in the Wild (AAAI 2025 · similarity 0.88)
- RepFace: Refining Closed-Set Noise with Progressive Label Correction for Face Recognition (AAAI 2025 · similarity 0.87)
- Open-Set 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)
- NoiseGPT: Label Noise Detection and Rectification through Probability Curvature (NeurIPS 2024 · similarity 0.84)
- Revisiting Interpolation for Noisy Label Correction (AAAI 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.
