Face Verification On Ijb C
評価指標
TAR @ FAR=1e-3
TAR @ FAR=1e-4
TAR @ FAR=1e-5
model
training dataset
評価結果
このベンチマークにおける各モデルのパフォーマンス結果
比較表
| モデル名 | TAR @ FAR=1e-3 | TAR @ FAR=1e-4 | TAR @ FAR=1e-5 | model | training dataset |
|---|---|---|---|---|---|
| circle-loss-a-unified-perspective-of-pair | 96.29% | 93.95% | 89.60% | R100 | MS1M Cleaned |
| elasticface-elastic-margin-loss-for-deep-face | - | 96.57% | - | R100 | MS1M V2 |
| magface-a-universal-representation-for-face | - | 95.97% | 90.36% | R100 | MS1MV2 |
| unified-negative-pair-generation-toward-well | - | - | 94.7% | - | - |
| arcface-additive-angular-margin-loss-for-deep | - | - | 96.07% | R100 | IBUG-500K |
| facenet-a-unified-embedding-for-face | - | - | - | - | - |
| killing-two-birds-with-one-stone-efficient | - | 97.97% | 96.93% | R200 | WebFace42M |
| rectifying-the-data-bias-in-knowledge | 97.05% | 95.49% | 93.25% | MobileFaceNet | MS1M V3 |
| cluster-and-aggregate-face-recognition-with | 98.08 | 97.3% | - | - | - |
| probabilistic-face-embeddings | 95.49% | - | - | SphereFace64 | MS1M V2 |
| an-efficient-training-approach-for-very-large | - | 97.31% | - | R100 | WebFace42M |
| controllable-and-guided-face-synthesis-for | - | 95.9% | 94.06% | - | - |
| it-s-all-in-the-head-representation-knowledge | - | 95.48% | 93.50% | MobileFaceNet | MS1M V3 |
| multicolumn-networks-for-face-recognition | - | - | - | - | - |
| adaface-quality-adaptive-margin-for-face | - | 97.39% | - | - | - |
| webface260m-a-benchmark-unveiling-the-power | - | 97.7% | - | R100 | WebFace42M |
| adaface-quality-adaptive-margin-for-face | - | 97.09% | - | - | - |
| curricularface-adaptive-curriculum-learning | - | 96.1% | - | - | - |
| qmagface-simple-and-accurate-quality-aware | 97.62 | 96.19% | - | - | - |
| unified-negative-pair-generation-toward-well | 97.57 | 96.38% | 94.47% | R100 | MS1MV2 |
| killing-two-birds-with-one-stone-efficient | - | 98.00% | 97.23% | ViT-L | WebFace42M |
| adaface-quality-adaptive-margin-for-face | - | 96.89% | - | - | - |
| look-across-elapse-disentangled | - | - | - | - | - |
| vggface2-a-dataset-for-recognising-faces | 92.7% | - | - | R50 | Vggface2 |
| it-s-all-in-the-head-representation-knowledge | - | 95.64% | 93.73% | MobileFaceNet | MS1M V3 |
| unified-negative-pair-generation-toward-well | 97.51 | 96.33% | - | - | - |