Infrastructure classification on FMOW WILDS (test)
74.8Average AccuracyWiSE-FT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| WiSE-FTalpha=optimal, fine-tuning strategy=end-to-end2021.09 | 74.8 | — | |
| CLIPmode=fine-tuned, fine-tuning strategy=end-to-end2021.09 | 73.3 | — | |
| WiSE-FTalpha=0.5, fine-tuning strategy=end-to-end2021.09 | 73 | — | |
| ERMBackbone=DenseNet-1212021.04 | 53.1 | 31.7 | |
| ERM (ours)Backbone=DenseNet-121, Implementation=ours2021.04 | 52.9 | 30.9 | |
| FishBackbone=DenseNet-1212021.04 | 51.8 | 34.3 | |
| IRMBackbone=DenseNet-1212021.04 | 50.9 | 31 | |
| CoralBackbone=DenseNet-1212021.04 | 50.5 | 30.5 | |
| CLIPmode=zero-shot2021.09 | 28 | — | |
| CLIPmode=zero-shot2021.09 | — | 23.8 | |
| CLIPmode=fine-tuned, fine-tuning strategy=end-to-end2021.09 | — | 46 | |
| WiSE-FTalpha=0.5, fine-tuning strategy=end-to-end2021.09 | — | 49.5 | |
| WiSE-FTalpha=optimal, fine-tuning strategy=end-to-end2021.09 | — | 49.7 |