Research map / Architectures and efficiency / Attention and transformers
TransFace: Calibrating Transformer Training for Face Recognition from a Data-Centric Perspective
Jun Dan, Yang Liu, Haoyu Xie, Jiankang Deng, Haoran Xie, Xuansong Xie, Baigui Sun
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 DPAP and EHSM (TransFace) | ViT-S · IJB-C TAR@FAR=1E-4 (MS1MV2) | TAR: 95.89 -> 96.45 | Improved |
| add DPAP and EHSM (TransFace) | ViT-L · IJB-C TAR@FAR=1E-4 (Glint360K) | TAR: 97.13 -> 97.61 | Improved |
| add DPAP and EHSM (TransFace) | ViT-L · IJB-C TAR@FAR=1E-5 (Glint360K) | TAR: 95.78 -> 96.29 | Improved |
| add DPAP and EHSM (TransFace) | ViT-L · CFP-FP (Glint360K) | accuracy: 99.10 -> 99.32 | Improved |
| add DPAP and EHSM (TransFace) | ViT-L · LFW (Glint360K) | accuracy: 99.82 -> 99.85 | No clear effect |
| add SE module (ablation) | ViT-S · IJB-C TAR@FAR=1E-4 (MS1MV2) | TAR: 95.89 -> 96.12 | Improved |
| add DPAP on top of SE (ablation) | ViT-S · IJB-C TAR@FAR=1E-4 (MS1MV2) | TAR: 96.12 -> 96.30 | Improved |
| add EHSM on top of DPAP (ablation) | ViT-S · IJB-C TAR@FAR=1E-4 (MS1MV2) | TAR: 96.30 -> 96.45 | Improved |
| use DPAP instead of Random Erasing (ablation) | ViT-S · IJB-C TAR@FAR=1E-6 (MS1MV2) | TAR: 83.68 -> 86.60 | Improved |
| use DPAP instead of Mixup/CutMix (ablation) | ViT-S · IJB-C TAR@FAR=1E-6 (MS1MV2) | TAR: 85.30 -> 86.60 | Improved |
| use DPAP instead of RandAugment (ablation) | ViT-S · IJB-C TAR@FAR=1E-4 (MS1MV2) | TAR: 96.14 -> 96.30 | Improved |
| use EHSM instead of Focal loss (ablation) | ViT-S · IJB-C TAR@FAR=1E-4 (MS1MV2) | TAR: 96.11 -> 96.22 | Improved |
Closest papers by meaning
- Vision Transformers Need Registers (ICLR 2024 · similarity 0.85)
- HyperFace: Generating Synthetic Face Recognition Datasets by Exploring Face Embedding Hypersphere (ICLR 2025 · similarity 0.85)
- GIF: Generative Inspiration for Face Recognition at Scale (CVPR 2025 · similarity 0.85)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · similarity 0.85)
- ScaleKD: Strong Vision Transformers Could Be Excellent Teachers (NeurIPS 2024 · similarity 0.85)
- Vision Transformers with Self-Distilled Registers (NeurIPS 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.
