mlsanityData to 7 Oct 2026

CosFace: Large Margin Cosine Loss for Deep Face Recognition

arXiv 1801.09414 ↗First posted 29 Jan 2018cs.CVcited by 968

Not on the map. Why?

Cited by 31 in Diffusion models, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

Face recognition has made extraordinary progress owing to the advancement of deep convolutional neural networks (CNNs). The central task of face recognition, including face verification and identification, involves face feature discrimination. However, the traditional softmax loss of deep CNNs usually lacks the power of discrimination. To address this problem, recently several loss functions such as center loss, large margin softmax loss, and angular softmax loss have been proposed. All these improved losses share the same idea: maximizing inter-class variance and minimizing intra-class variance. In this paper, we propose a novel loss function, namely large margin cosine loss (LMCL), to realize this idea from a different perspective. More specifically, we reformulate the softmax loss as a cosine loss by L2 normalizing both features and weight vectors to remove radial variations, based on which a cosine margin term is introduced to further maximize the decision margin in the angular space. As a result, minimum intra-class variance and maximum inter-class variance are achieved by virtue of normalization and cosine decision margin maximization. We refer to our model trained with LMCL as CosFace. Extensive experimental evaluations are conducted on the most popular public-domain face recognition datasets such as MegaFace Challenge, Youtube Faces (YTF) and Labeled Face in the Wild (LFW). We achieve the state-of-the-art performance on these benchmarks, which confirms the effectiveness of our proposed approach.

Key facts. Cited by 968 papers in the map (data through 7 Oct 2026). First posted on arXiv 29 Jan 2018.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 202; 2022: 184; 2023: 195; 2024: 157; 2025: 133; 2026: 97.

968 papers.

Sort:

By area:

  1. 2022AdaFace: Quality Adaptive Margin for Face Recognitioncited by 390
  2. 2021MagFace: A Universal Representation for Face Recognition and Quality Assessmentcited by 332
  3. 2021Generalizing Face Forgery Detection with High-frequency Featurescited by 261
  4. 2022Mitigating Neural Network Overconfidence with Logit Normalizationcited by 251
  5. 2021SimSwap: An Efficient Framework For High Fidelity Face Swappingcited by 243
  6. 2021WebFace260M: A Benchmark Unveiling the Power of Million-Scale Deep Face Recognitioncited by 221
  7. 2021FSCE: Few-Shot Object Detection via Contrastive Proposal Encodingcited by 168
  8. 2022Rethinking Visual Geo-localization for Large-Scale Applicationscited by 160
  9. 2021ElasticFace: Elastic Margin Loss for Deep Face Recognitioncited by 143
  10. 2022MegaPortraits: One-shot Megapixel Neural Head Avatarscited by 131
  11. 2022How to Exploit Hyperspherical Embeddings for Out-of-Distribution Detection?cited by 128
  12. 2022Frequency Domain Model Augmentation for Adversarial Attackcited by 128
38 more
  1. 2021Deformable ProtoPNet: An Interpretable Image Classifier Using Deformable Prototypescited by 120
  2. 2023DCFace: Synthetic Face Generation with Dual Condition Diffusion Modelcited by 116
  3. 2023EigenPlaces: Training Viewpoint Robust Models for Visual Place Recognitioncited by 109
  4. 2021SynFace: Face Recognition with Synthetic Datacited by 106
  5. 2022Protecting Celebrities from DeepFake with Identity Consistency Transformercited by 95
  6. 2022Neural Collapse: A Review on Modelling Principles and Generalizationcited by 95
  7. 2021LowKey: Leveraging Adversarial Attacks to Protect Social Media Users from Facial Recognitioncited by 92
  8. 2021Adversarial Sticker: A Stealthy Attack Method in the Physical Worldcited by 92
  9. 2022DigiFace-1M: 1 Million Digital Face Images for Face Recognitioncited by 92
  10. 2021ReSSL: Relational Self-Supervised Learning with Weak Augmentationcited by 83
  11. 2022Few-Shot Class-Incremental Learning from an Open-Set Perspectivecited by 83
  12. 2021Improving Transferability of Adversarial Patches on Face Recognition with Generative Modelscited by 80
  13. 2022Represent, Compare, and Learn: A Similarity-Aware Framework for Class-Agnostic Countingcited by 80
  14. 2022Are All Losses Created Equal: A Neural Collapse Perspectivecited by 78
  15. 2022DiffFace: Diffusion-based Face Swapping with Facial Guidancecited by 77
  16. 2021FASA: Feature Augmentation and Sampling Adaptation for Long-Tailed Instance Segmentationcited by 76
  17. 2022Improving Vision Transformers by Revisiting High-frequency Componentscited by 75
  18. 2022A Survey on Long-Tailed Visual Recognitioncited by 75
  19. 2023Long-tailed Visual Recognition via Gaussian Clouded Logit Adjustmentcited by 75
  20. 2022Killing Two Birds with One Stone:Efficient and Robust Training of Face Recognition CNNs by Partial FCcited by 73
  21. 2023IDiff-Face: Synthetic-based Face Recognition through Fizzy Identity-Conditioned Diffusion Modelscited by 72
  22. 2021CR-FIQA: Face Image Quality Assessment by Learning Sample Relative Classifiabilitycited by 68
  23. 2021SoK: How Robust is Image Classification Deep Neural Network Watermarking? (Extended Version)cited by 67
  24. 2022PatchNet: A Simple Face Anti-Spoofing Framework via Fine-Grained Patch Recognitioncited by 67
  25. 2022Fine-Grained Face Swapping via Regional GAN Inversioncited by 67
  26. 2023BlendFace: Re-designing Identity Encoders for Face-Swappingcited by 66
  27. 2024Deepfake Generation and Detection: A Benchmark and Surveycited by 66
  28. 2021PTTR: Relational 3D Point Cloud Object Tracking with Transformercited by 65
  29. 2024EMOPortraits: Emotion-enhanced Multimodal One-shot Head Avatarscited by 65
  30. 2023Boosting Adversarial Transferability by Block Shuffle and Rotationcited by 64
  31. 2021SDD-FIQA: Unsupervised Face Image Quality Assessment with Similarity Distribution Distancecited by 63
  32. 2022SFace: Privacy-friendly and Accurate Face Recognition using Synthetic Datacited by 63
  33. 2021RSG: A Simple but Effective Module for Learning Imbalanced Datasetscited by 61
  34. 2023Structure Invariant Transformation for better Adversarial Transferabilitycited by 61
  35. 2022Visual Recognition with Deep Nearest Centroidscited by 59
  36. 2021Evaluating Large-Vocabulary Object Detectors: The Devil is in the Detailscited by 57
  37. 2022High-resolution Face Swapping via Latent Semantics Disentanglementcited by 57
  38. 2021Distract Your Attention: Multi-head Cross Attention Network for Facial Expression Recognitioncited by 55

50 most cited of 968 All 968 (JSON)