Research map / Label-efficient and robust learning / Recognition and long-tail
Norface: Improving Facial Expression Analysis by Identity Normalization
Hanwei Liu, Rudong An, Zhimeng Zhang, Bowen Ma, Wei Zhang, Yan Song, Yujing Hu, Chen Wei, Yu Ding
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 identity normalization (normalized images as extra input) (ablation) | Norface classification network · BP4D AU detection | F1: 67.2 -> 69.3 | Improved |
| add identity normalization (ablation) | Norface · RAF-DB | accuracy: 89.24 -> 92.97 | Improved |
| add identity normalization (ablation) | Norface · AffectNet | accuracy: 67.29 -> 68.69 | Improved |
| use identity normalization instead of mixup/cutout/cutmix augmentation (ablation) | Norface · BP4D AU detection | F1: 67.4 (CutMix) -> 69.3 | Improved |
| standard augmentation (mixup, cutout, cutmix) vs random init (ablation) | Norface baseline · BP4D+ AU detection | F1: 64.0 -> 64.2 to 64.8 | No clear effect |
| feed expression features instead of normalized images to classifier (ablation) | Norface · BP4D AU detection | F1: 69.3 -> 67.1 | Hurt |
| add Input and Output MoE modules (ablation) | Norface classification network · BP4D AU detection | F1: 65.9 -> 69.3 | Improved |
| add Input MoE (vs only Output MoE) (ablation) | Norface · RAF-DB | accuracy: 92.57 -> 92.97 | Improved |
| add importance loss (ablation) | Norface MoE · RAF-DB | accuracy: 91.66 -> 92.97 | Improved |
| add eyebrow loss to expression loss in normalization network (ablation) | normalization network · AU detection (eyebrow regions) | F1: +0.5% average | Improved |
| increase number of experts up to m=4 (ablation) | Norface MoE · AU detection and FER | F1 / accuracy: +3.0% and +5.74% vs m=0 | Improved |
| choose a poor target identity (ID3 vs ID6) (ablation) | Idn target face · BP4D AU detection | F1: -1.1% | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LA-Net — LA-Net: Landmark-Aware Learning for Reliable Facial Expression Recognition under Label Noise (ICCV 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- EmotioNet: An Accurate, Real-Time Algorithm for the Automatic Annotation of a Million Facial Expressions in the Wild (CVPR 2016 · similarity 0.88)
- Knowledge Augmented Deep Neural Networks for Joint Facial Expression and Action Unit Recognition (NeurIPS 2020 · similarity 0.86)
- Navigating Label Ambiguity for Facial Expression Recognition in the Wild (AAAI 2025 · similarity 0.85)
- Facial-R1: Aligning Reasoning and Recognition for Facial Emotion Analysis (AAAI 2026 · similarity 0.85)
- Open-Set Facial Expression Recognition (AAAI 2024 · similarity 0.85)
- AUEditNet: Dual-Branch Facial Action Unit Intensity Manipulation with Implicit Disentanglement (CVPR 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.
