Image Classification on ImageNet-ES Diverse new (test)
92.8Oracle Score (S)DINOv2-g
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DINOv2-gNum. Params=1.1B, Pretraining Dataset=LVD-142M, DG method=Dataset curation2025.03 | 92.8 | 82.5 | 62.8 | 35.3 | 82.9 | |
| OpenCLIP-hNum. Params=632M, Pretraining Dataset=LAION-2B, DG method=Text-guided pretrain2025.03 | 87.9 | 74.6 | 45.5 | 29.3 | 74.4 | |
| DINOv2-bNum. Params=90M, Pretraining Dataset=LVD-142M, DG method=Dataset curation2025.03 | 87.5 | 72.6 | 44.5 | 28.3 | 72.9 | |
| OpenCLIP-bNum. Params=87M, Pretraining Dataset=LAION-2B, DG method=Text-guided pretrain2025.03 | 82.7 | 66.4 | 38.8 | 24.5 | 67.6 | |
| ResNet-50Num. Params=26M, Pretraining Dataset=IN-21K, DG method=DeepAugment*+AugMix†2025.03 | 80.2 | 64.1 | 36.2 | 23.6 | 65.1 | |
| EfficientNet-B3Num. Params=12M, Pretraining Dataset=IN-1K, DG method=-2025.03 | 75.8 | 57.2 | 33.6 | 21.4 | 55.7 | |
| SwinV2-SNum. Params=50M, Pretraining Dataset=IN-1K, DG method=-2025.03 | 75.3 | 53.9 | 30.8 | 18.9 | 55.6 | |
| SwinV2-BNum. Params=88M, Pretraining Dataset=IN-1K, DG method=-2025.03 | 74.6 | 53.6 | 30.8 | 18.5 | 55.3 | |
| SwinV2-TNum. Params=28M, Pretraining Dataset=IN-1K, DG method=-2025.03 | 71.5 | 50.8 | 26.5 | 16.9 | 50.3 | |
| ResNet-152Num. Params=60M, Pretraining Dataset=IN-1K, DG method=-2025.03 | 69.2 | 43.4 | 21.9 | 14.2 | 48.8 | |
| EfficientNet-B0Num. Params=5M, Pretraining Dataset=IN-1K, DG method=-2025.03 | 66.9 | 42.2 | 21.8 | 14 | 45.9 | |
| ResNet-50Num. Params=26M, Pretraining Dataset=IN-1K, DG method=-2025.03 | 63.1 | 38.2 | 17.6 | 12 | 43.3 |