Multi-task Image Classification on 30 Vision Tasks
93.1AccuracyIndividual
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| IndividualBackbone=ViT-B/322026.06 | 93.1 | — | |
| ReTeXBackbone=ViT-B/32, Merging Type=Dynamic model merging2026.06 | 91.2 | 97.9 | |
| WEMoEBackbone=ViT-B/32, Merging Type=Dynamic model merging2026.06 | 67.1 | 72.4 | |
| Consensus TABackbone=ViT-B/32, Merging Type=Static model merging2026.06 | 63.4 | 68.5 | |
| Twin-MergingBackbone=ViT-B/32, Merging Type=Dynamic model merging2026.06 | 60.1 | 65.2 | |
| TIES-MergingBackbone=ViT-B/32, Merging Type=Static model merging2026.06 | 59.6 | 64.7 | |
| Weight AveragingBackbone=ViT-B/32, Merging Type=Static model merging2026.06 | 59.1 | 64.2 | |
| Task ArithmeticBackbone=ViT-B/32, Merging Type=Static model merging2026.06 | 58 | 62.8 | |
| AdaMergingBackbone=ViT-B/32, Merging Type=Static model merging2026.06 | 56.8 | 61.3 | |
| MoW-MergingBackbone=ViT-B/32, Merging Type=Dynamic model merging2026.06 | 56.4 | 60.6 | |
| Zero-shotBackbone=ViT-B/322026.06 | 55.5 | — | |
| DaWinBackbone=ViT-B/32, Merging Type=Dynamic model merging2026.06 | 40.3 | 42.9 |