Image Classification on Zero-Shot Evaluation Suite (14 Datasets) (test val)
76ImageNet-1k AccuracyLiT
Evaluation results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| LiTPre-training Dataset=IN-21k, Model Scale=L, Training Data=Unfiltered WebLI, Evaluation Protocol=Zero-shot Classification2023.05 | 76 | 82.9 | 82.4 | 87.1 | 51.8 | 62 | 45.6 | 66.1 | 67.2 | 41.9 | 27.6 | 72.6 | 29 | 65.4 | 68.3 | |
| 3TPre-training Dataset=IN-21k, Model Scale=L, Training Data=Unfiltered WebLI, Evaluation Protocol=Zero-shot Classification2023.05 | 71.7 | 73.4 | 84.1 | 87 | 60.3 | 51.8 | 54.3 | 88.1 | 64.9 | 58.3 | 42.8 | 65.7 | 57.9 | 68.7 | 71.4 | |
| BaselinePre-training Dataset=IN-21k, Model Scale=L, Training Data=Unfiltered WebLI, Evaluation Protocol=Zero-shot Classification2023.05 | 69.5 | 73.5 | 81.9 | 84.2 | 58.6 | 49.6 | 53 | 85.8 | 62.2 | 56.2 | 32.7 | 62 | 58 | 67.6 | 68.4 |