Research map / Image generation / Editing and style transfer
SynFER: Towards Boosting Facial Expression Recognition with Synthetic Data
Xilin He, Cheng Luo, Xiaole Xian, Bing Li, Muhammad Haris Khan, Zongyuan Ge, Weicheng Xie, Siyang Song, Linlin Shen, Bernard Ghanem, Xiangyu Yue
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 FAU control to the text-conditioned generator (ablation) | SynFER generator (FEText) · Generation FER accuracy | FER accuracy: 34.62 -> 48.74 | Improved |
| add FAU control to the text-conditioned generator (ablation) | SynFER generator (FEText) · Generation FAU detection | FAU accuracy: 88.91 -> 92.37 | Improved |
| add semantic guidance on top of FEText+FAUs (ablation) | SynFER generator · Generation FER accuracy | FER accuracy: 48.74 -> 55.14 | Improved |
| add semantic guidance on top of FEText+FAUs (ablation) | SynFER generator · AffectNet supervised (POSTER++) | accuracy: 67.68 -> 68.13 | Improved |
| fine-tune Stable Diffusion on FEText (ablation) | Stable Diffusion · AffectNet supervised (POSTER++ on synthetic data) | accuracy: 65.36 -> 66.62 | Improved |
| add synthetic data to real AffectNet data for self-supervised pretraining | MoCo v3 · AffectNet linear probe | accuracy: 51.03 -> 57.84 | Improved |
| pretrain on synthetic data alone instead of real AffectNet | MoCo v3 · RAF-DB linear probe | accuracy: 79.05 -> 81.17 | Improved |
| add synthetic data to real AffectNet data for pretraining | SimCLR · AffectNet linear probe | accuracy: 48.36 -> 54.37 | Improved |
| add synthetic data to real AffectNet data for pretraining | BYOL · SFEW linear probe | accuracy: 48.70 -> 51.70 | Improved |
| add synthetic data to supervised training | POSTER++ · AffectNet | accuracy: 67.49 -> 69.04 | Improved |
| add synthetic data to supervised training | APViT · RAF-DB and AffectNet | accuracy: +0.27% on RAF-DB and 0.32% on AffectNet | Improved |
| add synthetic data to real data for few-shot training | CDNet · RAF_C 5-way 1-shot | accuracy: 46.07 -> 46.97 | Improved |
Closest papers by meaning
- Masked Linear Regression for Learning Local Receptive Fields for Facial Expression Synthesis (IJCV 2019 · similarity 0.87)
- FERV39k: A Large-Scale Multi-Scene Dataset for Facial Expression Recognition in Videos (CVPR 2022 · similarity 0.87)
- SynFace: Face Recognition With Synthetic Data (ICCV 2021 · similarity 0.87)
- D2SP: Dynamic Dual-Stage Purification Framework for Dual Noise Mitigation in Vision-based Affective Recognition. (CVPR 2025 · similarity 0.87)
- Facial Expression Recognition with Inconsistently Annotated Datasets (ECCV 2018 · similarity 0.87)
- Suppressing Uncertainties for Large-Scale Facial Expression Recognition (CVPR 2020 · similarity 0.87)
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.
