Research map / Label-efficient and robust learning / Recognition and long-tail
AdaptiveFace: Adaptive Margin and Sampling for Face Recognition
Check 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 |
|---|---|---|---|
| replace CosFace's uniform margin with learned class-specific margins (ablation) | CosFace face recognition model · MegaFace | MF1 Veri.: 94.115 -> 95.032 | Helped |
| add Adaptive Data Sampling (ablation) | CosFace face recognition model · MegaFace | MF1 Rank 1: 93.942 -> 94.068 | Helped |
| add Hard Prototype Mining (ablation) | CosFace face recognition model · MegaFace | MF1 Veri.: 94.115 -> 94.789 | Helped |
| combine ADS, HPM, and AdaM-Softmax (ablation) | CosFace face recognition model · MegaFace | MF1 Veri.: 94.115 -> 95.608 | Helped |
| set AdaM-Softmax margin constraint λ to 0 (ablation) | AdaM-Softmax · MegaFace | MF1 Veri.: 95.032 -> 93.348 | Hurt |
| increase AdaM-Softmax λ from 50 to 70 (ablation) | AdaM-Softmax · MegaFace | MF1 Rank 1: 94.373 -> 94.590 | Helped |
| increase AdaM-Softmax λ from 70 to 100 (ablation) | AdaM-Softmax · MegaFace | MF1 Veri.: 94.687 -> 94.641 | Hurt |
| set HPM similarity threshold h to 0.15 (ablation) | CosFace with Hard Prototype Mining · LFW | verification accuracy: 99.53 -> 99.62 | Helped |
| set HPM similarity threshold h to 0.23 (ablation) | CosFace with Hard Prototype Mining · LFW BLUFR | VR@FAR=0.001%: 98.27 -> 97.72 | Hurt |
| use AdaptiveFace instead of CosFace | 50-layer ResNet trained for face recognition · MegaFace Challenge 1, FaceScrub probe | MF1 Rank1: 93.942 -> 95.023 | Helped |
| use AdaptiveFace instead of CosFace | 50-layer ResNet trained for face recognition · LFW | verification accuracy: 99.53 -> 99.62 | Helped |
| use AdaptiveFace instead of ArcFace | 50-layer ResNet trained for face recognition · LFW BLUFR | VR@FAR=0.001%: 98.48 -> 98.89 | Helped |
Closest papers by meaning
- AdaFace: Quality Adaptive Margin for Face Recognition (CVPR 2022 · similarity 0.92)
- UniTSFace: Unified Threshold Integrated Sample-to-Sample Loss for Face Recognition (NeurIPS 2023 · similarity 0.89)
- RepFace: Refining Closed-Set Noise with Progressive Label Correction for Face Recognition (AAAI 2025 · similarity 0.87)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · similarity 0.86)
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (AAAI 2025 · similarity 0.86)
- HyperFace: Generating Synthetic Face Recognition Datasets by Exploring Face Embedding Hypersphere (ICLR 2025 · similarity 0.86)
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.