Research map / Label-efficient and robust learning / Recognition and long-tail
Rotation Consistent Margin Loss for Efficient Low-Bit Face Recognition
Yudong Wu, Yichao Wu, Ruihao Gong, Yuanhao Lv, Ken Chen, Ding Liang, Xiaolin Hu, Xianglong Liu, Junjie Yan
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 RCM loss to low-bit ResNet18 | 4-bit ResNet18 face recognition · MegaFace, 1M distractors | rank-1 identification accuracy: 87.74 -> 88.56 | Improved |
| Add RCM loss to low-bit ResNet18 | 3-bit ResNet18 face recognition · MegaFace, 1M distractors | rank-1 identification accuracy: 72.84 -> 74.34 | Improved |
| Add RCM loss to MobileNetV2 | 4-bit MobileNetV2 face recognition · MegaFace, 1M distractors | rank-1 identification accuracy: 77.90 -> 80.12 | Improved |
| Add RCM loss to MobileNetV2 | 3-bit MobileNetV2 face recognition · MegaFace, 1M distractors | rank-1 identification accuracy: 59.59 -> 62.58 | Improved |
| Add RCM loss instead of ArcFace+Mimic | 4-bit ResNet18 · LFW | verification accuracy: 98.68 -> 98.91 | Improved |
| Add RCM loss instead of ArcFace+Mimic | 4-bit ResNet18 · YTF | verification accuracy: 94.76 -> 94.95 | Improved |
| Add feature mimic supervision to ArcFace (ablation) | 4-bit ResNet18 · MegaFace, 1M distractors | rank-1 identification accuracy: 67.75 -> 69.04 | Improved |
| Replace ArcFace+Mimic with RCM loss (ablation) | 4-bit ResNet18 · MegaFace, 1M distractors | rank-1 identification accuracy: 69.04 -> 71.57 | Improved |
| Use RCM angular-margin formula LA rather than LB or LC (ablation) | 4-bit ResNet18 · MegaFace | accuracy | Improved |
| Set the RCM scaling parameter to λ = 5.0 rather than other tested values (ablation) | 4-bit ResNet18 with LA · MegaFace | accuracy | Mixed |
| Add RCM loss to Dorefa-Net quantization (ablation) | 4-bit ResNet18 · MegaFace, 1M distractors | rank-1 identification accuracy: 90.59 -> 91.55 | Improved |
| Add RCM loss to DSQ quantization (ablation) | 4-bit ResNet18 · MegaFace, 1M distractors | rank-1 identification accuracy: 88.98 -> 89.46 | Improved |
Closest papers by meaning
- ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019 · similarity 0.87)
- CosFace: Large Margin Cosine Loss for Deep Face Recognition (CVPR 2018 · similarity 0.87)
- Mis-Classified Vector Guided Softmax Loss for Face Recognition (AAAI 2020 · similarity 0.86)
- UniTSFace: Unified Threshold Integrated Sample-to-Sample Loss for Face Recognition (NeurIPS 2023 · similarity 0.86)
- More Information Supervised Probabilistic Deep Face Embedding Learning (ICML 2020 · similarity 0.86)
- Switchable Token-Specific Codebook Quantization For Face Image Compression (NeurIPS 2025 · 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.
