Research map / Label-efficient and robust learning / Recognition and long-tail
TransFace++: Rethinking the Face Recognition Paradigm With a Focus on Accuracy, Efficiency, and Security
Jun Dan, Yang Liu, Baigui Sun, Jiankang Deng, Shan Luo
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 TransFace (DPAP and EHSM) to ViT-B | ViT-B · IJB-C TAR@FAR=1e-5 (WebFace42M, RGB) | verification accuracy (%): 96.07 -> 97.25 | Improved |
| add TransFace to ViT-L | ViT-L · IJB-C TAR@FAR=1e-4 (WebFace42M, RGB) | verification accuracy (%): 97.65 -> 98.16 | Improved |
| add TransFace to ViT-S on small MS1MV2 | ViT-S · IJB-C TAR@FAR=1e-4 (MS1MV2, RGB) | verification accuracy (%): 95.89 -> 96.45 | Improved |
| train TransFace on limited-size MS1MV2 versus large sets | TransFace · MS1MV2 IJB-B/IJB-C | verification accuracy (%) | Mixed |
| add TIBC and SICA to ViT-L on PNG bytes | ViT-L · IJB-C TAR@FAR=1e-4 (MS1MV2, PNG bytes) | verification accuracy (%): 93.75 -> 96.37 | Improved |
| add TIBC and SICA to ViT-L on TIFF bytes | ViT-L · IJB-C TAR@FAR=1e-4 (MS1MV2, TIFF bytes) | verification accuracy (%): 95.25 -> 96.85 | Improved |
| feed TIFF bytes with TransFace++-L instead of RGB ViT-L | TransFace++-L · IJB-C TAR@FAR=1e-4 (MS1MV2) | verification accuracy (%): +0.61% | Improved |
| use PNG instead of TIFF encoding for bytes input | TransFace++-L · IJB-C TAR@FAR=1e-4 (MS1MV2) | verification accuracy (%): 96.85 -> 96.37 | Hurt |
| randomly shuffle byte order during training (ablation) | ViT-S TIFF bytes · IJB-C TAR@FAR=1e-4 (MS1MV2) | verification accuracy (%): 94.74 -> 74.25 | Hurt |
| apply strided sampling to bytes (ablation) | ViT-S TIFF bytes · IJB-C TAR@FAR=1e-4 (MS1MV2) | verification accuracy (%): 94.74 -> 79.56 | Hurt |
| apply window shuffle to bytes (ablation) | ViT-S TIFF bytes · IJB-C TAR@FAR=1e-4 (MS1MV2) | verification accuracy (%): 94.74 -> 83.30 | Hurt |
| reverse byte order (ablation) | ViT-S TIFF bytes · IJB-C TAR@FAR=1e-4 (MS1MV2) | verification accuracy (%): 94.74 -> 93.76 | Mixed |
Closest papers by meaning
- TransFace: Calibrating Transformer Training for Face Recognition from a Data-Centric Perspective (ICCV 2023 · similarity 0.90)
- LVFace: Progressive Cluster Optimization for Large Vision Models in Face Recognition (ICCV 2025 · similarity 0.87)
- WebFace260M: A Benchmark for Million-Scale Deep Face Recognition (TPAMI 2022 · similarity 0.87)
- TopoFR: A Closer Look at Topology Alignment on Face Recognition (NeurIPS 2024 · similarity 0.85)
- FSFM: A Generalizable Face Security Foundation Model via Self-Supervised Facial Representation Learning (CVPR 2025 · 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.
