Face Recognition on VGGFace2 (test)
98.21Accte_rm Accuracyretrained
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| retrainedBackbone=ResNet50, Drm=true, Dfg=false, RTE=5870.35, # param. (%)=25,557,032 (100%)2026.01 | 98.21 | 0 | 0 | 0 | |
| pretrainedBackbone=ResNet50, Drm=true, Dfg=true, RTE=5970.95, # param. (%)=25,557,032 (100%)2026.01 | 98.15 | 99.49 | 100 | — | |
| SalUnBackbone=ResNet50, Drm=true, Dfg=true, RTE=312.71, learning rate=5 × 10−2, # param. (%)=12,778,516 (50%)2026.01 | 97 | 0 | 0 | 0.25 | |
| FTBackbone=ResNet50, Drm=true, Dfg=false, RTE=294.62, learning rate=5 × 10−2, # param. (%)=25,557,032 (100%)2026.01 | 96.22 | 0 | 0 | 0.43 | |
| SalUnBackbone=ResNet50, Drm=false, Dfg=true, RTE=9.99, learning rate=10−4, # param. (%)=12,778,516 (50%)2026.01 | 95.59 | 5.08 | 0 | 2.7 | |
| RLBackbone=ResNet50, Drm=false, Dfg=true, RTE=5.59, learning rate=5 × 10−4, # param. (%)=25,557,032 (100%)2026.01 | 95.27 | 13.2 | 0 | 5.55 | |
| DELETEBackbone=ResNet50, Drm=false, Dfg=true, RTE=5.82, learning rate=5 × 10−4, # param. (%)=25,557,032 (100%)2026.01 | 95.15 | 5.58 | 0 | 2.93 | |
| BTBackbone=ResNet50, Drm=true, Dfg=true, RTE=8.24, learning rate=5 × 10−4, # param. (%)=25,557,032 (100%)2026.01 | 94.97 | 4.06 | 0 | 2.58 | |
| RLBackbone=ResNet50, Drm=true, Dfg=true, RTE=303.81, learning rate=5 × 10−2, # param. (%)=25,557,032 (100%)2026.01 | 94.66 | 0 | 0 | 0.89 | |
| LOFTBackbone=ResNet50, Drm=only Σrm pre, Dfg=only Σfg pre, RTE=0.45, learning rate=1, steps=50, # param. (%)=43,008 (0.17%)2026.01 | 94.22 | 0.51 | 0 | 1.46 | |
| L2ULBackbone=ResNet50, Drm=false, Dfg=true, RTE=176.69, learning rate=1 × 10−4, # param. (%)=25,557,032 (100%)2026.01 | 92.88 | 3.55 | 0 | 2.71 | |
| GABackbone=ResNet50, Drm=false, Dfg=true, RTE=3.35, learning rate=5 × 10−4, # param. (%)=25,557,032 (100%)2026.01 | 86.34 | 0 | 0 | 4 |