| It's All in the Head: Representation Knowledge Distillation through Classifier Sharing | ✓ Link | 90.24% | 93.73% | 95.64% | | | MS1M V3 | MobileFaceNet | | | HeadSharing: SH-KD | 2022-01-18 |
| It's All in the Head: Representation Knowledge Distillation through Classifier Sharing | ✓ Link | 89.82% | 93.50% | 95.48% | | | MS1M V3 | MobileFaceNet | | | HeadSharing: TH-KD | 2022-01-18 |
| Controllable and Guided Face Synthesis for Unconstrained Face Recognition | ✓ Link | 89.34% | 94.06% | 95.9% | | | | | 96.31 | 97.48 | ArcFace+CSFM | 2022-07-20 |
| Killing Two Birds with One Stone:Efficient and Robust Training of Face Recognition CNNs by Partial FC | ✓ Link | | 97.23% | 98.00% | | | WebFace42M | ViT-L | | | Partial FC | 2022-03-28 |
| Killing Two Birds with One Stone:Efficient and Robust Training of Face Recognition CNNs by Partial FC | ✓ Link | | 96.93% | 97.97% | | | WebFace42M | R200 | | | PartialFC | 2022-03-28 |
| ArcFace: Additive Angular Margin Loss for Deep Face Recognition | ✓ Link | | 96.07% | | | | IBUG-500K | R100 | | | ArcFace | 2018-01-23 |
| Unified Negative Pair Generation toward Well-discriminative Feature Space for Face Recognition | ✓ Link | | 94.7% | | | | | | | | Mag+UNPG | 2022-03-22 |
| Unified Negative Pair Generation toward Well-discriminative Feature Space for Face Recognition | ✓ Link | | 94.47% | 96.38% | 97.57 | | MS1MV2 | R100 | | | Cos+UNPG | 2022-03-22 |
| Rectifying the Data Bias in Knowledge Distillation | | | 93.25% | 95.49% | 97.05% | | MS1M V3 | MobileFaceNet | | | L2E+IS-sampling | 2021-10-11 |
| MagFace: A Universal Representation for Face Recognition and Quality Assessment | ✓ Link | | 90.36% | 95.97% | | | MS1MV2 | R100 | | | MagFace++ | 2021-03-11 |
| Circle Loss: A Unified Perspective of Pair Similarity Optimization | ✓ Link | | 89.60% | 93.95% | 96.29% | | MS1M Cleaned | R100 | | | circle loss | 2020-02-25 |
| WebFace260M: A Benchmark Unveiling the Power of Million-Scale Deep Face Recognition | | | | 97.7% | | | WebFace42M | R100 | | | WebFace42M baseline | 2021-03-06 |
| AdaFace: Quality Adaptive Margin for Face Recognition | ✓ Link | | | 97.39% | | | | | | | AdaFace (WebFace4M) | 2022-04-03 |
| An Efficient Training Approach for Very Large Scale Face Recognition | ✓ Link | | | 97.31% | | | WebFace42M | R100 | | | FFC | 2021-05-21 |
| Cluster and Aggregate: Face Recognition with Large Probe Set | ✓ Link | | | 97.3% | 98.08 | | | | | | CAFace+AdaFace (WebFace4M) | 2022-10-19 |
| AdaFace: Quality Adaptive Margin for Face Recognition | ✓ Link | | | 97.09% | | | | | | | AdaFace (MS1MV3) | 2022-04-03 |
| AdaFace: Quality Adaptive Margin for Face Recognition | ✓ Link | | | 96.89% | | | | | | | AdaFace (MS1MV2) | 2022-04-03 |
| ElasticFace: Elastic Margin Loss for Deep Face Recognition | ✓ Link | | | 96.57% | | | MS1M V2 | R100 | | | ElasticFace-Cos | 2021-09-20 |
| Unified Negative Pair Generation toward Well-discriminative Feature Space for Face Recognition | ✓ Link | | | 96.33% | 97.51 | | | | | | Arc+UNPG | 2022-03-22 |
| QMagFace: Simple and Accurate Quality-Aware Face Recognition | ✓ Link | | | 96.19% | 97.62 | 98.51 | | | | | QMagFace | 2021-11-26 |
| CurricularFace: Adaptive Curriculum Learning Loss for Deep Face Recognition | ✓ Link | | | 96.1% | | | | | | | CurricularFace | 2020-04-01 |
| Probabilistic Face Embeddings | ✓ Link | | | | 95.49% | 97.17% | MS1M V2 | SphereFace64 | | | PFEfuse + match | 2019-04-21 |
| VGGFace2: A dataset for recognising faces across pose and age | ✓ Link | | | | 92.7% | 96.7% | Vggface2 | R50 | | | VGGFace2_ft | 2017-10-23 |
| Look Across Elapse: Disentangled Representation Learning and Photorealistic Cross-Age Face Synthesis for Age-Invariant Face Recognition | ✓ Link | | | | | 93.5% | | | | | AIM | 2018-09-02 |
| Multicolumn Networks for Face Recognition | ✓ Link | | | | | 92.70% | | | | | MN-vc | 2018-07-24 |
| FaceNet: A Unified Embedding for Face Recognition and Clustering | ✓ Link | | | | | 66.5% | | | | | FaceNet | 2015-03-12 |
| Vision Transformers for Face Recognition Need More Registers | ✓ Link | | | 97.37% | | | | | | | ViT-8R | 2026-06-10 |
| LVFace: Large Vision model for Face Recogniton | ✓ Link | | | 98.06% | | | | | | | LVFace-L (WebFace42M) | 2025-01-23 |
| FaceLiVT: Face Recognition using Linear Vision Transformer with Structural Reparameterization For Mobile Device | | | | 95.7% | | | | | | | FaceLiVT-M-(LA, 40 epochs) | 2025-06-12 |
| EdgeFace: Efficient Face Recognition Model for Edge Devices | ✓ Link | | | 95.63% | | | | | | | EdgeFace-S (gamma=0.5) | 2023-07-04 |
| LAFS: Landmark-based Facial Self-supervised Learning for Face Recognition | ✓ Link | | | 97.48% | | | | | | | Part fViT, ours (LAFS) | 2024-03-13 |
| Sparsity-Inducing Divergence Losses for Biometric Verification | | | | 97.78% | | | | | | | Q-Margin | 2026-06-30 |
| Learning to Attract and Repel: Dual Quality Margin Learning for Face Recognition (DQM-Face) | | | | 96.58% | | | | | | | DQM-Face α=0.4 | 2026-09-08 |
| Noise-Robust Face Recognition via Non-target Similarity Distribution Guided Sample Selection | | | | 96.96% | | | | | | | Arc+DDLN (R100) | 2026-09-08 |
| Partial FC: Training 10 Million Identities on a Single Machine | ✓ Link | | | 97.3% | | | | | | | Partial FC ResNet100 + CosFace (Glint360K, r=1.0) | 2020-10-11 |
| SwinFace: A Multi-task Transformer for Face Recognition, Expression Recognition, Age Estimation and Attribute Estimation | ✓ Link | | | 96.73% | | | | | | | SwinFace | 2023-08-22 |
| TransFace: Calibrating Transformer Training for Face Recognition from a Data-Centric Perspective | ✓ Link | | | 98.16% | | | | | | | TransFace-L (WebFace42M) | 2023-08-20 |
| KeyPoint Relative Position Encoding for Face Recognition | ✓ Link | | | 97.82% | | | | | | | AdaFace ViT Base KP-RPE (WebFace12M) | 2024-03-21 |
| ARoFace: Alignment Robustness to Improve Low-Quality Face Recognition | ✓ Link | | | 97.6% | | | | | 97.61 | 98.57 | AdaFace+ARoFace (WebFace12M) | 2024-07-20 |
| SFace: Sigmoid-Constrained Hypersphere Loss for Robust Face Recognition | ✓ Link | | | 96.11% | | | | | | | SFace (MS1MV2, ResNet100) | 2022-05-24 |