Image Classification Accuracy (Top-1) on ImageNet (train)
75.3Top-1 AccuracySwAV
Evaluation results
| Method | Links | |
|---|---|---|
| SwAVarchitecture=R50-MLP (28M), pre-training epochs=800, evaluation augmentation=multi-crop, batch size=40962020.05 | 75.3 | |
| BYOLarchitecture=R50-MLPhid(35M), pre-training epochs=1000, batch size=40962020.05 | 74.3 | |
| SimCLRarchitecture=R50-MLP (28M), pre-training epochs=1000, batch size=40962020.05 | 69.3 | |
| AMDIMarchitecture=Custom (626M), pre-training epochs=150, evaluation protocol=FastAutoAugment2020.05 | 68.1 | |
| PCL v2architecture=R50-MLP (28M), pre-training epochs=2002020.05 | 67.6 | |
| MoCo v2architecture=R50-MLP (28M), pre-training epochs=2002020.05 | 67.5 | |
| CMCarchitecture=R50L+ab (47M), pre-training epochs=2802020.05 | 64 | |
| PIRLarchitecture=R50 (24M), pre-training epochs=8002020.05 | 63.6 | |
| SimCLRarchitecture=R50-MLP (28M), pre-training epochs=2002020.05 | 61.9 | |
| PCLarchitecture=R50 (24M), pre-training epochs=2002020.05 | 61.5 | |
| SelfLabelarchitecture=R50 (24M), pre-training epochs=4002020.05 | 61.5 | |
| MoCoarchitecture=R50 (24M), pre-training epochs=2002020.05 | 60.6 | |
| LocalAggarchitecture=R50 (24M), pre-training epochs=200, evaluation augmentation=10-crop2020.05 | 60.2 | |
| BigBiGANarchitecture=R50 (24M)2020.05 | 56.6 | |
| InstDiscarchitecture=R50 (24M), pre-training epochs=2002020.05 | 54 | |
| Rotationarchitecture=R50 (24M)2020.05 | 48.9 | |
| CPCarchitecture=R101 (28M)2020.05 | 48.7 | |
| DeepClusterarchitecture=VGG(15M), pre-training epochs=1002020.05 | 48.4 | |
| Jigsawarchitecture=R50 (24M), pre-training epochs=902020.05 | 45.7 |