Image Classification on Eurosat (Acc, Rob)
79.31AccuracyGEOM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GEOMMeta-Album Size=Extended, Training Dataset=Meta-Album2025.07 | 79.31 | — | |
| GEOMMeta-Album Size=Mini, Training Dataset=Meta-Album2025.07 | 78.4 | — | |
| GEOMMeta-Album Size=Micro, Training Dataset=Meta-Album2025.07 | 63.7 | — | |
| GEOM-INTraining Dataset=ImageNet-1k2025.07 | 56.7 | — | |
| TPTSetting=Zero-shot, Backbone=ViT-B/16, Attack=PGD-100, Perturbation Budget (epsilon_a)=4/255, Patch Granularity=16x162026.05 | 42.6 | 0 | |
| MTASetting=Zero-shot, Backbone=ViT-B/16, Attack=PGD-100, Perturbation Budget (epsilon_a)=4/255, Patch Granularity=16x162026.05 | 42.5 | 1.2 | |
| C-TPTSetting=Zero-shot, Backbone=ViT-B/16, Attack=PGD-100, Perturbation Budget (epsilon_a)=4/255, Patch Granularity=16x162026.05 | 42.3 | 0 | |
| CLIPSetting=Zero-shot, Backbone=ViT-B/16, Attack=PGD-100, Perturbation Budget (epsilon_a)=4/255, Patch Granularity=16x162026.05 | 42.2 | 0 | |
| R-TPTSetting=Zero-shot, Backbone=ViT-B/16, Attack=PGD-100, Perturbation Budget (epsilon_a)=4/255, Patch Granularity=16x162026.05 | 34.7 | 8.5 | |
| CLBPSetting=Zero-shot, Backbone=ViT-B/16, Attack=PGD-100, Perturbation Budget (epsilon_a)=4/255, Patch Granularity=16x162026.05 | 34 | 16.6 |