Research Article
A Community Detection Approach to Cleaning Extremely Large Face Database
Table 3
Comparison with other state-of-the-art results on the LFW benchmark (performance in %).
| | | #Net | Pairwise verification | Close-set identification | Open-set identification | | Acc | VR@FAR = 0.1% | Rank 1 | DIR@FAR = 1% | DIR@FAR = 0.1% |
| | DeepID2+ | 25 | 99.47 | - | 95.00 | 80.70 | - | | Face++ | 4 | 99.50 | - | - | - | - | | DeepID3 | 25 | 99.53 | - | 96.00 | 81.40 | - | | IDL Single Model | 7 | 99.68 | 99.11 | 97.60 | 94.12 | 89.08 | | IDL Ensemble Model | 70 | 99.77 | 99.41 | 98.03 | 95.80 | 92.09 |
| | DeepFace | 1 | 97.35 | - | 64.90 | 44.50 | - | | WebFace | 1 | 97.73 | 80.26 | - | - | 28.90 | | Facebook-WST Fusion | 1 | 98.37 | - | 82.50 | 61.90 | - | | CenterLoss | 1 | 98.70 | - | 94.05 | 69.97 | - | | Light CNN-29 | 1 | 99.33 | - | 97.33 | 93.62 | - | | FaceNet | 1 | 99.63 | - | - | - | - | | Ours | 1 | 99.67 | 99.53 | 98.31 | 95.40 | 86.31 |
|
|
