Image Classification on ImageNet 1k (val) (Accuracy & TPU Efficiency)
90Top-1 AccuracyMPL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MPLPre-training dataset=unlabelled or weakly labelled data (proprietary), Resolution=448, Evaluation protocol=Fine-tuned2021.05 | 90 | — | 20,480 | |
| NFNet-F4+Pre-training dataset=JFT-300M (proprietary), Resolution=448, Evaluation protocol=Fine-tuned2021.05 | 89.2 | 46 | 1,860 | |
| ALIGNPre-training dataset=unlabelled or weakly labelled data (proprietary), Resolution=448, Evaluation protocol=Fine-tuned2021.05 | 88.64 | 15 | 14,820 | |
| ViT-H/14Pre-training dataset=JFT-300M (proprietary), Resolution=448, Evaluation protocol=Fine-tuned2021.05 | 88.55 | 15 | 2,300 | |
| Mixer-H/14Pre-training dataset=JFT-300M (proprietary), Resolution=448, Evaluation protocol=Fine-tuned2021.05 | 87.94 | 40 | 1,010 | |
| BiT-R152x4Pre-training dataset=JFT-300M (proprietary), Resolution=448, Evaluation protocol=Fine-tuned2021.05 | 87.54 | 26 | 9,900 | |
| HaloNetPre-training dataset=ImageNet-21k (public), Resolution=448, Evaluation protocol=Fine-tuned2021.05 | 85.8 | 120 | 100 | |
| HaloNetPre-training dataset=ImageNet-21k (public), Resolution=384, Evaluation protocol=Fine-tuned2021.05 | 85.5 | 258 | — | |
| BiT-R152x4Pre-training dataset=ImageNet-21k (public), Resolution=448, Evaluation protocol=Fine-tuned2021.05 | 85.39 | 26 | 940 | |
| ViT-L/16Pre-training dataset=ImageNet-21k (public), Resolution=448, Evaluation protocol=Fine-tuned2021.05 | 85.3 | 32 | 180 | |
| ViT-L/16Pre-training dataset=ImageNet-21k (public), Resolution=384, Evaluation protocol=Fine-tuned2021.05 | 85.15 | 80 | — | |
| Mixer-L/16Pre-training dataset=ImageNet-21k (public), Resolution=448, Evaluation protocol=Fine-tuned2021.05 | 84.15 | 105 | 410 | |
| Mixer-L/16Pre-training dataset=ImageNet-21k (public), Resolution=224, Evaluation protocol=Fine-tuned2021.05 | 82.84 | 420 | — |