Zero-shot Classification on EuroSAT
68.5Top-1 Clean AccuracyVanilla FT
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Vanilla FTBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 68.5 | — | — | — | 0 | — | — | |
| GRACEBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 67.2 | — | — | — | 31.5 | — | — | |
| TPGMBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 66.1 | — | — | — | 0 | — | — | |
| SPDBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 65.3 | — | — | — | 0 | — | — | |
| FLYPBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 64.8 | — | — | — | 0 | — | — | |
| WISE-FTBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 62.4 | — | — | — | 0 | — | — | |
| CLIPBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 54.7 | — | — | — | 0 | — | — | |
| FAREBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 48.6 | — | — | — | 14.5 | — | — | |
| TTC2026.01 | 45.6 | — | 10.2 | — | — | — | — | |
| PMG-AFTBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 44.2 | — | — | — | 27.8 | — | — | |
| CLIP2026.01 | 42.9 | — | 0 | — | — | — | — | |
| CSR2026.01 | 42.9 | — | 40.4 | — | — | — | — | |
| TTE2026.01 | 42.3 | — | 8.3 | — | — | — | — | |
| HD2026.01 | 41.7 | — | 8.5 | — | — | — | — | |
| LPF2026.01 | 41.6 | — | 4.9 | — | — | — | — | |
| LAATBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 41.3 | — | — | — | 25.6 | — | — | |
| Anti-Adv2026.01 | 38.7 | — | 25.5 | — | — | — | — | |
| TeCoABackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 38.6 | — | — | — | 23.48 | — | — | |
| R-TPT2026.01 | 29.6 | — | 22.3 | — | — | — | — | |
| TeCoA2026.01 | 16.5 | — | 16.4 | — | — | — | — | |
| FARE2026.01 | 12.6 | — | 12.6 | — | — | — | — | |
| CLIPBackbone=ViT-B/16, Prompting Strategy=Standard, Training Samples per Class=0, Training Protocol=Zero-shot (Pre-trained)2024.01 | — | 64.05 | — | — | — | — | — | |
| CLIP + ABackbone=ViT-B/16, Prompting Strategy=LLM Attributes, Training Samples per Class=0, Training Protocol=Zero-shot (Pre-trained)2024.01 | — | 71.51 | — | — | — | — | — | |
| CLIP-A-selfBackbone=ViT-B/16, Prompting Strategy=GPT Attributes, Training Samples per Class=16, Training Protocol=Adapter Tuning2024.01 | — | 70.5 | — | — | — | — | — | |
| CLIP-BaselineParadigm=GVLM, Tuned on=-, Backbone=ViT-B-322023.06 | — | 47.21 | — | — | — | — | — | |
| CLIPFT + ABackbone=ViT-B/16, Prompting Strategy=LLM Attributes, Training Samples per Class=16, Training Protocol=Fine-tuned2024.01 | — | 81.82 | — | — | — | — | — | |
| GeoRSCLIPParadigm=GVLM + Pfeiffer, Tuned on=RS5M, Backbone=ViT-B-322023.06 | — | 61.41 | — | — | — | — | — | |
| GeoRSCLIPParadigm=GVLM + Prefixtuning, Tuned on=RS5M, Backbone=ViT-B-322023.06 | — | 61.59 | — | — | — | — | — | |
| GeoRSCLIPParadigm=GVLM + LORA, Tuned on=RS5M, Backbone=ViT-B-322023.06 | — | 60.1 | — | — | — | — | — | |
| GeoRSCLIPParadigm=GVLM + UniPELT, Tuned on=RS5M, Backbone=ViT-B-322023.06 | — | 60.46 | — | — | — | — | — | |
| GeoRSCLIPParadigm=GVLM + FT, Tuned on=RS5M, Backbone=ViT-B-322023.06 | — | 61.49 | — | — | — | — | — | |
| GeoRSCLIPParadigm=GVLM + FT, Tuned on=RS5M, Backbone=ViT-H-142023.06 | — | 67.47 | — | — | — | — | — | |
| TAMEBackbone=ViT-B/32, Evaluation Protocol=Zero-shot, Prompt Configuration=V-L Independent2026.05 | — | — | — | 22.2 | 31.5 | 20.6 | 28.9 | |
| VanillaBackbone=ViT-B/32, Evaluation Protocol=Zero-shot2026.05 | — | — | — | 0 | 0.1 | 0 | 0.7 |