Research map / Label-efficient and robust learning / Recognition and long-tail
Facial Expression Recognition with Inconsistently Annotated Datasets
Jiabei Zeng, Shiguang Shan, Xilin Chen
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 |
|---|---|---|---|
| replace probability transition layers with a category-general fully connected layer (e2e-fc) (ablation) | IPA2LT(LTNet) · FER overall average | accuracy: 70.64 -> 28.87 | Hurt |
| add probability transition layers (end-to-end latent truth) instead of two-step EM then CNN (ablation) | IPA2LT · FER overall average | accuracy: 68.53 -> 70.64 | Improved |
| use end-to-end LTNet instead of EM+CNN (spectral init) (ablation) | LTNet · CIFAR-10 noisy labels | accuracy: 78.82 -> 87.23 | Improved |
| use all three inconsistent label sets instead of a single noisy set | basic CNN · CIFAR-10 noisy labels | accuracy: 62.50 -> 84.04 | Improved |
| model each coder's annotation bias with LTNet instead of mixing labels in a CNN (ablation) | basic CNN on mixture · CIFAR-10 noisy labels | accuracy: 84.04 -> 87.23 | Improved |
| use LTNet instead of NAL | IPA2LT · FER overall average | accuracy: 70.15 -> 70.64 | Improved |
| use LTNet instead of AIR | IPA2LT · FER overall average | accuracy: 55.21 -> 70.64 | Improved |
| add LTNet with predicted annotations from unlabeled data to AffTr+RAFTr training | base ResNet-80 · FER overall average | accuracy: 69.77 -> 70.64 | Improved |
| LTNet vs NAL on AffectNet | IPA2LT · AffectNet val | accuracy: 55.97 -> 55.11 | Mixed |
Closest papers by meaning
- Understanding and Mitigating Annotation Bias in Facial Expression Recognition (ICCV 2021 · similarity 0.90)
- Navigating Label Ambiguity for Facial Expression Recognition in the Wild (AAAI 2025 · similarity 0.88)
- FAIR-FER: A Latent Alignment Approach for Mitigating Bias in Facial Expression Recognition (Student Abstract) (AAAI 2024 · similarity 0.88)
- Open-Set Facial Expression Recognition (AAAI 2024 · similarity 0.88)
- Facial-R1: Aligning Reasoning and Recognition for Facial Emotion Analysis (AAAI 2026 · similarity 0.87)
- D2SP: Dynamic Dual-Stage Purification Framework for Dual Noise Mitigation in Vision-based Affective Recognition. (CVPR 2025 · 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.
