Research map / Label-efficient and robust learning / Recognition and long-tail
AnchorFace: Boosting TAR@FAR for Practical Face Recognition
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 AnchorFace losses to CosFace | CosFace · IJB-B | TAR@FAR=1e-4: 94.20 -> 94.38 | Helped |
| Add AnchorFace losses to ArcFace | ArcFace · IJB-B | TAR@FAR=1e-4: 94.25 -> 94.42 | Helped |
| Add AnchorFace losses to CurricularFace | CurricularFace · IJB-B | TAR@FAR=1e-4: 94.85 -> 94.97 | Helped |
| Add AnchorFace losses to CosFace | CosFace · IJB-C | TAR@FAR=1e-4: 95.85 -> 96.09 | Helped |
| Add AnchorFace losses to ArcFace | ArcFace · IJB-C | TAR@FAR=1e-4: 95.91 -> 96.22 | Helped |
| Add AnchorFace losses to CurricularFace | CurricularFace · IJB-C | TAR@FAR=1e-4: 96.13 -> 96.32 | Helped |
| Add AnchorFace losses to ArcFace | ArcFace · IFRT (African, Caucasian, Indian, Asian, and All) | TAR@FAR=1e-6: 79.228 -> 79.314; 86.718 -> 87.001; 85.405 -> 85.593; 58.341 -> 59.702; 81.235 -> 82.062 | Helped |
| Remove TAR loss (ablation) | AF-ArcFace · IJB-C | TAR@FAR=1e-4: 96.22 -> 96.02 | Hurt |
| Remove FAR loss (ablation) | AF-ArcFace · IJB-C | TAR@FAR=1e-4: 96.22 -> 96.05 | Hurt |
| Replace the online-updating set with classifier weights and current-batch features (ablation) | AF-ArcFace · IJB-C | TAR@FAR=1e-4: 96.22 -> 95.98 | Hurt |
| Vary maximum stored features per identity K from 1 to 3 to 5 to 7 to 9 (ablation) | AF-ArcFace · IJB-C | TAR@FAR=1e-4: 94.28; 95.01; 95.23; 95.22; 95.25 | Mixed |
| Vary maximum valid steps M from 400 to 800 to 1000 to 1200 to 1400 (ablation) | AF-ArcFace · IJB-C | TAR@FAR=1e-4: 95.09; 95.16; 95.23; 95.20; 95.18 | Mixed |
Closest papers by meaning
- UniTSFace: Unified Threshold Integrated Sample-to-Sample Loss for Face Recognition (NeurIPS 2023 · similarity 0.87)
- Seeing Small Faces From Robust Anchor's Perspective (CVPR 2018 · similarity 0.86)
- WebFace260M: A Benchmark Unveiling the Power of Million-Scale Deep Face Recognition (CVPR 2021 · similarity 0.86)
- ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019 · similarity 0.85)
- TopoFR: A Closer Look at Topology Alignment on Face Recognition (NeurIPS 2024 · similarity 0.83)
- GIF: Generative Inspiration for Face Recognition at Scale (CVPR 2025 · similarity 0.83)
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.