Image Classification on Seen Tasks (EuroSAT, SVHN, GTSRB, MNIST) (test)
97.13EuroSAT AccuracyBD-Merging
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BD-MergingBackbone=ViT-B/32, Weights=Layer-wise2026.03 | 97.13 | 89.2 | 92.58 | 99.17 | 94.53 | |
| Twin-MergingBackbone=ViT-B/32, Weights=Layer-wise2026.03 | 94.83 | 87.1 | 91.6 | 98.72 | 93.06 | |
| AdaMergingBackbone=ViT-B/32, Weights=Layer-wise2026.03 | 89.85 | 88.14 | 87.31 | 97.94 | 90.79 | |
| Task ArithmeticBackbone=ViT-B/32, Weights=Layer-wise2026.03 | 75.8 | 86.69 | 78.77 | 97.8 | 84.77 | |
| Ties-MergingBackbone=ViT-B/32, Weights=Layer-wise2026.03 | 70.85 | 82.47 | 70.43 | 96.14 | 79.97 | |
| PretrainedBackbone=ViT-B/322026.03 | 43.6 | 32.62 | 32.6 | 48.26 | 39.27 |