Image Classification on ImageNet (val) (Accuracy and MAdds)
76.4AccuracySupervised
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SupervisedEpochs=90, Evaluation Protocol=Linear Evaluation2021.03 | 76.4 | — | |
| SwAV + BG_RMEpochs=800, Evaluation Protocol=Linear Evaluation2021.03 | 76.1 | — | |
| SwAV + BG_RandomEpochs=800, Evaluation Protocol=Linear Evaluation2021.03 | 76.1 | — | |
| SwAVEpochs=800, Evaluation Protocol=Linear Evaluation2021.03 | 75.3 | — | |
| SwAV (repro.)Epochs=800, Evaluation Protocol=Linear Evaluation2021.03 | 74.9 | — | |
| BYOL + BG_RandomEpochs=1000, Evaluation Protocol=Linear Evaluation2021.03 | 74.8 | — | |
| BYOL + BG_RMEpochs=1000, Evaluation Protocol=Linear Evaluation2021.03 | 74.6 | — | |
| BYOLEpochs=1000, Evaluation Protocol=Linear Evaluation2021.03 | 74.3 | — | |
| TuNASSearch space=S1, Search algorithm=TuNAS, Lines of codes=+1, Search cost=1, Train cost=12021.01 | 74.2 | 301 | |
| Bayesian SearchSearch space=S1, Search algorithm=Bayesian, Lines of codes=+1, Search cost=25, Train cost=12021.01 | 73.9 | 301 | |
| BYOL + BG_RandomEpochs=300, Evaluation Protocol=Linear Evaluation2021.03 | 73.9 | — | |
| TuNASSearch space=S3, Search algorithm=TuNAS, Lines of codes=+1, Search cost=2, Train cost=12021.01 | 73.8 | 302 | |
| BYOL (repro.)Epochs=1000, Evaluation Protocol=Linear Evaluation2021.03 | 73.8 | — | |
| Random SearchSearch space=S1, Search algorithm=RS, Lines of codes=+23, Search cost=25, Train cost=12021.01 | 73.7 | 300 | |
| SwAV + BG_RMEpochs=100, Evaluation Protocol=Linear Evaluation2021.03 | 73.6 | — | |
| Cyclic Learning Rate Restarting (CLR)Model=ResNet-34, Version=A, Pruning=l1-norm based filter pruning2021.05 | 73.44 | — | |
| SwAV + BG_RandomEpochs=100, Evaluation Protocol=Linear Evaluation2021.03 | 73.4 | — | |
| TuNASSearch space=S2, Search algorithm=TuNAS, Lines of codes=+10, Search cost=1, Train cost=12021.01 | 73.3 | 302 | |
| Unpruned BaselineModel=ResNet-342021.05 | 73.3 | — | |
| BYOL + BG_RMEpochs=300, Evaluation Protocol=Linear Evaluation2021.03 | 73.3 | — | |
| Barlow TwinsEpochs=1000, Evaluation Protocol=Linear Evaluation2021.03 | 73.2 | — | |
| Cyclic Learning Rate Restarting (CLR)Model=ResNet-34, Version=B, Pruning=l1-norm based filter pruning2021.05 | 73.14 | — | |
| MobileNetV2 (static baseline)Search space=(static), Search algorithm=N/A, Lines of codes=N/A, Search cost=N/A, Train cost=12021.01 | 73.1 | 300 | |
| Scratch-BModel=ResNet-34, Version=A, Pruning=l1-norm based filter pruning2021.05 | 73.03 | — | |
| Fine-tuningModel=ResNet-34, Version=A, Pruning=l1-norm based filter pruning2021.05 | 72.96 | — | |
| Scratch-BModel=ResNet-34, Version=B, Pruning=l1-norm based filter pruning2021.05 | 72.91 | — | |
| BYOL (repro.)Epochs=300, Evaluation Protocol=Linear Evaluation2021.03 | 72.7 | — | |
| Scratch-EModel=ResNet-34, Version=A, Pruning=l1-norm based filter pruning2021.05 | 72.56 | — | |
| Fine-tuningModel=ResNet-34, Version=B, Pruning=l1-norm based filter pruning2021.05 | 72.5 | — | |
| BYOLEpochs=300, Evaluation Protocol=Linear Evaluation2021.03 | 72.5 | — | |
| Scratch-EModel=ResNet-34, Version=B, Pruning=l1-norm based filter pruning2021.05 | 72.29 | — | |
| SwAV (repro.)Epochs=100, Evaluation Protocol=Linear Evaluation2021.03 | 72.2 | — | |
| MoCo-v2 + BG_SwapsEpochs=800, Evaluation Protocol=Linear Evaluation2021.03 | 72.2 | — | |
| SwAVEpochs=100, Evaluation Protocol=Linear Evaluation2021.03 | 72.1 | — | |
| MoCo-v2 + BG_RMEpochs=800, Evaluation Protocol=Linear Evaluation2021.03 | 71.9 | — | |
| MoCo-v2Epochs=800, Evaluation Protocol=Linear Evaluation2021.03 | 71.1 | — | |
| MoCo-v2 (repro.)Epochs=800, Evaluation Protocol=Linear Evaluation2021.03 | 71 | — | |
| MoCo-v2 + BG_SwapsEpochs=220, Evaluation Protocol=Linear Evaluation2021.03 | 69.5 | — | |
| SimCLREpochs=1000, Evaluation Protocol=Linear Evaluation2021.03 | 69.3 | — | |
| MoCo-v2 + BG_RMEpochs=220, Evaluation Protocol=Linear Evaluation2021.03 | 69.1 | — | |
| DiLo (MoCo-v2)Epochs=200, Evaluation Protocol=Linear Evaluation2021.03 | 67.9 | — | |
| MoCo-v2 (repro.)Epochs=220, Evaluation Protocol=Linear Evaluation2021.03 | 67.7 | — | |
| PCL-v2Epochs=200, Evaluation Protocol=Linear Evaluation2021.03 | 67.6 | — | |
| MoCo-v2Epochs=200, Evaluation Protocol=Linear Evaluation2021.03 | 67.5 | — | |
| SimCLREpochs=200, Evaluation Protocol=Linear Evaluation2021.03 | 66.8 | — | |
| CMCEpochs=200, Evaluation Protocol=Linear Evaluation2021.03 | 66.2 | — | |
| PIRLEpochs=800, Evaluation Protocol=Linear Evaluation2021.03 | 63.6 | — | |
| SeLaEpochs=400, Evaluation Protocol=Linear Evaluation2021.03 | 61.5 | — | |
| MoCoEpochs=200, Evaluation Protocol=Linear Evaluation2021.03 | 60.6 | — |