Research map / Label-efficient and robust learning / Recognition and long-tail
Stylized-Face: A Million-level Stylized Face Dataset for Face Recognition
Zhengyuan Peng, Jianqing Xu, Yuge Huang, Jinkun Hao, Shouhong Ding, Zhizhong Zhang, Xin Tan, Lizhuang Ma
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 |
|---|---|---|---|
| train on Stylized-Face instead of MS1MV3 | IResNet-50 face recognition · Standard stylized face recognition | TAR@FAR=1e-4: 49.7 -> 90.6 | Improved |
| train on Stylized-Face instead of Glint360k | IResNet-50 face recognition · Standard stylized face recognition | TAR@FAR=1e-4: 74.7 -> 90.6 | Improved |
| train on Stylized-Face instead of Glint360k | IResNet-50 face recognition · Cross-prompt stylized face recognition | TAR@FAR=1e-4: 73.9 -> 90.3 | Improved |
| train on Stylized-Face instead of Glint360k | IResNet-50 face recognition · Cross-method (all generators) | TAR@FAR=1e-3: 58.6 -> 71.9 | Improved |
| train on Stylized-Face instead of Glint360k | IResNet-50 face recognition · Cross-method Flashface hard prompts | TAR@FAR=1e-2: 84.9 -> 87.1 | Improved |
| train on Stylized-Face instead of Glint360k | IResNet-50 face recognition · Cross-method Flashface easy prompts | TAR@FAR=1e-2: 96.9 -> 96.9 | No clear effect |
| scale training data from MS1MV3 to larger Glint360k | IResNet-50 face recognition · Standard stylized face recognition | TAR@FAR=1e-4: 49.7 -> 74.7 | Improved |
| use IResNet-101 instead of IResNet-50 backbone (ablation) | Stylized-Face training · Standard stylized face recognition | TAR@FAR=1e-4: 90.6 -> 92.0 | Improved |
| use Arcface instead of Cosface (ablation) | Stylized-Face training · Standard stylized face recognition | TAR@FAR=1e-4: 90.6 -> 89.9 | No clear effect |
| use CurricularFace instead of Cosface (ablation) | Stylized-Face training · Standard stylized face recognition | TAR@FAR=1e-4: 90.6 -> 90.5 | No clear effect |
| train on Stylized-Face instead of Glint360k | IResNet-50 face recognition · Real face recognition average | accuracy: 99.1 -> 98.7 | Hurt |
Closest papers by meaning
- Data Synthesis with Diverse Styles for Face Recognition via 3DMM-Guided Diffusion (CVPR 2025 · similarity 0.86)
- $\text{ID}^3$: Identity-Preserving-yet-Diversified Diffusion Models for Synthetic Face Recognition (NeurIPS 2024 · similarity 0.86)
- UIFace: Unleashing Inherent Model Capabilities to Enhance Intra-Class Diversity in Synthetic Face Recognition (ICLR 2025 · similarity 0.86)
- HyperFace: Generating Synthetic Face Recognition Datasets by Exploring Face Embedding Hypersphere (ICLR 2025 · similarity 0.86)
- CemiFace: Center-based Semi-hard Synthetic Face Generation for Face Recognition (NeurIPS 2024 · similarity 0.85)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 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.
