Research map / Label-efficient and robust learning / Recognition and long-tail
CosFace: Large Margin Cosine Loss for Deep 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 a cosine margin to the normalized softmax objective | 64-layer CNN trained with CASIA-WebFace · YTF | accuracy (%) | Helped |
| add a cosine margin to the normalized softmax objective | 64-layer CNN trained with CASIA-WebFace · LFW | accuracy (%) | Mixed |
| add feature normalization (ablation) | CosFace with LMCL · LFW | accuracy (%): 99.10 -> 99.33 | Helped |
| add feature normalization (ablation) | CosFace with LMCL · YTF | accuracy (%): 93.1 -> 96.1 | Helped |
| add feature normalization (ablation) | CosFace with LMCL · MegaFace Challenge 1 | Rank 1 (%): 75.10 -> 77.11 | Helped |
| add feature normalization (ablation) | CosFace with LMCL · MegaFace Challenge 1 | verification TAR (%) at 10^-6 FAR: 88.65 -> 89.88 | Helped |
| increase cosine margin m from zero toward 0.35 (ablation) | CosFace trained on CASIA-WebFace · LFW and YTF | accuracy (%) | Helped |
| increase cosine margin beyond the convergent range (ablation) | CosFace trained on CASIA-WebFace · training | convergence | Hurt |
| use a 3-patch ensemble instead of a single patch (ablation) | CosFace · MegaFace Challenge 1, Small protocol | Rank 1 (%): 77.11 -> 79.54 | Helped |
| use a 3-patch ensemble instead of a single patch (ablation) | CosFace · MegaFace Challenge 2, Large protocol | Rank 1 (%): 74.11 -> 77.06 | Helped |
Closest papers by meaning
- AdaCos: Adaptively Scaling Cosine Logits for Effectively Learning Deep Face Representations (CVPR 2019 · similarity 0.91)
- Mis-Classified Vector Guided Softmax Loss for Face Recognition (AAAI 2020 · similarity 0.91)
- ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019 · similarity 0.90)
- Killing Two Birds With One Stone: Efficient and Robust Training of Face Recognition CNNs by Partial FC (CVPR 2022 · similarity 0.89)
- P2SGrad: Refined Gradients for Optimizing Deep Face Models (CVPR 2019 · similarity 0.89)
- UniTSFace: Unified Threshold Integrated Sample-to-Sample Loss for Face Recognition (NeurIPS 2023 · similarity 0.88)
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.