Image Classification Worst-Task Accuracy on eight-task benchmark CLIP-ViT
76.3Worst-task Top-1 AccuracyEpiMer
Evaluation results
| Method | Links | |
|---|---|---|
| EpiMerBackbone=ViT-L/142026.05 | 76.3 | |
| TSV-MBackbone=ViT-L/142026.05 | 76.2 | |
| EpiMerBackbone=ViT-B/162026.05 | 71 | |
| TSV-MBackbone=ViT-B/162026.05 | 68.6 | |
| TIESBackbone=ViT-L/142026.05 | 67.5 | |
| EpiMerBackbone=ViT-B/322026.05 | 66.5 | |
| TSV-MBackbone=ViT-B/322026.05 | 65.7 | |
| AM / TABackbone=ViT-L/142026.05 | 61.9 | |
| TIESBackbone=ViT-B/162026.05 | 55.9 | |
| TIESBackbone=ViT-B/322026.05 | 53.2 | |
| FisherBackbone=ViT-L/142026.05 | 50.7 | |
| AM / TABackbone=ViT-B/322026.05 | 50.6 | |
| AM / TABackbone=ViT-B/162026.05 | 50.4 | |
| FisherBackbone=ViT-B/162026.05 | 40.8 | |
| FisherBackbone=ViT-B/322026.05 | 33.8 |