Research map / Label-efficient and robust learning / Recognition and long-tail
Circle Loss: A Unified Perspective of Pair Similarity Optimization
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 ArcFace with CircleLoss | ResNet34 face recognition · MegaFace Challenge 1, Rank 1 | Rank 1 (%): 97.68 -> 97.81 | Helped |
| Replace ArcFace with CircleLoss | ResNet34 face verification · IJB-C 1:1, TAR@FAR 1e-5 | TAR (%): 84.23 -> 86.78 | Helped |
| Replace AM-Softmax with CircleLoss | MGN re-identification model · Market-1501 | R-1 (%): 95.3 -> 96.1 | Helped |
| Replace AM-Softmax with CircleLoss | MGN re-identification model · MSMT17 | R-1 (%): 76.5 -> 76.9 | Helped |
| Use CircleLoss instead of the strongest listed retrieval baseline | 512-D retrieval model · Cars196 | R@1 (%): 85.2 -> 83.4 | Hurt |
| Use CircleLoss instead of Multi-Simi | 512-D retrieval model · CUB-200-2011 | R@1 (%): 65.7 -> 66.7 | Helped |
| Use CircleLoss instead of Multi-Simi | 512-D retrieval model · Stanford Online Products | R@1 (%): 78.2 -> 78.3 | Helped |
| Add MGN part-level features to the ResNet50 CircleLoss model (ablation) | CircleLoss re-identification model · Market-1501 | R-1 (%): 94.2 -> 96.1 | Helped |
| Add MGN part-level features to the ResNet50 CircleLoss model (ablation) | CircleLoss re-identification model · MSMT17 | R-1 (%): 76.3 -> 76.9 | Helped |
| Vary scale factor gamma from 32 to 1024 (ablation) | CircleLoss face recognition · Face recognition | performance: 32 to 1024 | No clear effect |
| Vary relaxation margin m across the reported settings (ablation) | CircleLoss face recognition · Face recognition | performance: -0.05 to 0.25 | No clear effect |
Closest papers by meaning
- Improved Deep Metric Learning with Multi-class N-pair Loss Objective (NeurIPS 2016 · similarity 0.87)
- Multi-Similarity Loss With General Pair Weighting for Deep Metric Learning (CVPR 2019 · similarity 0.86)
- Ring Loss: Convex Feature Normalization for Face Recognition (CVPR 2018 · similarity 0.86)
- OLÉ: Orthogonal Low-Rank Embedding - A Plug and Play Geometric Loss for Deep Learning (CVPR 2018 · similarity 0.86)
- Stabilizing Sample Similarity in Representation via Mitigating Random Consistency (ICML 2025 · similarity 0.85)
- Cross-Domain Similarity Learning for Face Recognition in Unseen Domains (CVPR 2021 · similarity 0.85)
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.