Research map / Label-efficient and robust learning / Recognition and long-tail
Ring Loss: Convex Feature Normalization for 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 Ring loss to Softmax | Softmax, Res64 · MegaFace, 1 million distractors | identification rate (%): 56.36 -> 71.67 | Helped |
| Add Ring loss to SphereFace | SphereFace, Res64 · IJB-A Janus 1:1, $10^{-5}$ FAR | verification rate (%): 78.52% to 82.41% | Helped |
| Add Ring loss to SphereFace | SphereFace, Res64 · Janus CS3 1:1, $10^{-4}$ FAR | verification rate (%): 82.29 -> 82.74 | Helped |
| Add Ring loss to Softmax | Softmax, Res64 · CFP Frontal vs. Profile, $10^{-3}$ FAR | verification rate (%): 55.86 -> 87.43 | Helped |
| Add Ring loss to Softmax | Softmax, Res64 · Janus CS3 downsampling by factor 64x | verification performance | Helped |
| Use Ring loss weight $0.0001$ instead of $0.01$ (ablation) | Softmax + Ring loss · LFW | accuracy (%): 99.52 -> 99.28 | Hurt |
| Use Ring loss weight $0.03$ instead of $0.01$ (ablation) | SphereFace + Ring loss · IJB-A Janus 1:1, $10^{-5}$ FAR | verification rate (%): 82.41 -> 79.54 | Hurt |
| Vary Ring loss weight across reported settings (ablation) | Softmax + Ring loss · Janus CS3 1:1, $10^{-4}$ FAR | verification rate (%): 73.56; 74.56; 71.06 | Mixed |
| Vary Ring loss weight across reported settings (ablation) | SphereFace + Ring loss · Janus CS3 1:1, $10^{-4}$ FAR | verification rate (%): 76.97; 82.74; 81.62; 81.57 | Mixed |
| Replace soft feature normalization with hard projection using $l_2$-constrained Softmax | Softmax feature normalization · Janus CS3 downsampling by factor 64x | verification performance | Hurt |
| Add Ring loss to Softmax | Softmax · MNIST | accuracy (%): 98.97 -> 99.34 | Helped |
Closest papers by meaning
- CosFace: Large Margin Cosine Loss for Deep Face Recognition (CVPR 2018 · similarity 0.87)
- Circle Loss: A Unified Perspective of Pair Similarity Optimization (CVPR 2020 · similarity 0.86)
- Learning to Regress 3D Face Shape and Expression From an Image Without 3D Supervision (CVPR 2019 · similarity 0.85)
- Mis-Classified Vector Guided Softmax Loss for Face Recognition (AAAI 2020 · similarity 0.85)
- RegularFace: Deep Face Recognition via Exclusive Regularization (CVPR 2019 · similarity 0.84)
- ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019 · similarity 0.84)
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.