Image Classification Robustness on ImageNet-C mean over shifts v1 (test)
83.7AccuracyViT-22B
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ViT-22BFine-tuning=true, Resolution=3842023.02 | 83.7 | 0.63 | 1 | 97 | |
| ViT-22BResolution=384, Fine-tuning=ImageNet2023.02 | 83.7 | 0.63 | 1 | 97 | |
| Plex-22B [BE]Fine-tuning=true, Resolution=384, Heads=BatchEnsemble2023.02 | 81 | 0.98 | 18 | 95 | |
| Plex-22B [HET]Fine-tuning=true, Resolution=384, Heads=Heteroscedastic2023.02 | 80.9 | 0.97 | 17 | 94 | |
| Plex-L/32Fine-tuning=true, Resolution=3842023.02 | 71.3 | 1.21 | 2 | 91 | |
| Plex-L/32Resolution=384, Fine-tuning=ImageNet, Efficient-ensemble=true, Heteroscedastic layers=true2023.02 | 71.3 | 1.21 | 2 | 91 | |
| ViT-L/32Fine-tuning=true, Resolution=3842023.02 | 70.1 | 1.28 | 5 | 91 | |
| ViT-L/32Resolution=384, Fine-tuning=ImageNet2023.02 | 70.1 | 1.28 | 5 | 91 |