Multi-task Image Classification on 8 Vision Tasks Norm Mismatch setting (test)
93.3Absolute AccuracyIndividual
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| IndividualBackbone=ViT-L-142025.08 | 93.3 | — | |
| EMR-Merging + DisTaCBackbone=ViT-L-142025.08 | 92.7 | 99 | |
| WUDI-Merging + DisTaCBackbone=ViT-L-142025.08 | 91.4 | 97.5 | |
| Iso-CTSBackbone=ViT-L-142025.08 | 90.8 | 96.9 | |
| TSVM + DisTaCBackbone=ViT-L-142025.08 | 90.3 | 96.6 | |
| Iso-CTS + DisTaCBackbone=ViT-L-142025.08 | 90.1 | 96.1 | |
| IndividualBackbone=ViT-B-322025.08 | 89.3 | — | |
| EMR-Merging + DisTaCBackbone=ViT-B-322025.08 | 88.1 | 97.3 | |
| EMR-MergingBackbone=ViT-L-142025.08 | 87.6 | 93.6 | |
| Consensus TA + DisTaCBackbone=ViT-L-142025.08 | 84.9 | 90.7 | |
| TSVMBackbone=ViT-L-142025.08 | 84.8 | 90.7 | |
| TIES + DisTaCBackbone=ViT-L-142025.08 | 84.4 | 90.2 | |
| WUDI-Merging + DisTaCBackbone=ViT-B-322025.08 | 84.4 | 93.2 | |
| Task arithmetic + DisTaCBackbone=ViT-L-142025.08 | 83.9 | 89.6 | |
| TSVM + DisTaCBackbone=ViT-B-322025.08 | 82.9 | 91.8 | |
| Consensus TABackbone=ViT-L-142025.08 | 82 | 87.6 | |
| Iso-CTS + DisTaCBackbone=ViT-B-322025.08 | 80.3 | 88.9 | |
| EMR-MergingBackbone=ViT-B-322025.08 | 80 | 88.7 | |
| Task arithmeticBackbone=ViT-L-142025.08 | 78.6 | 84.2 | |
| Iso-CTSBackbone=ViT-B-322025.08 | 78.1 | 86.2 | |
| TIESBackbone=ViT-L-142025.08 | 74 | 79.5 | |
| Consensus TA + DisTaCBackbone=ViT-B-322025.08 | 73.7 | 82.2 | |
| TIES + DisTaCBackbone=ViT-B-322025.08 | 73.1 | 81 | |
| TSVMBackbone=ViT-B-322025.08 | 72.2 | 80.2 | |
| Task arithmetic + DisTaCBackbone=ViT-B-322025.08 | 70 | 78.2 | |
| Consensus TABackbone=ViT-B-322025.08 | 68.8 | 77 | |
| Pre-trainedBackbone=ViT-L-142025.08 | 65.1 | — | |
| Task arithmeticBackbone=ViT-B-322025.08 | 63.6 | 71.8 | |
| TIESBackbone=ViT-B-322025.08 | 59.1 | 66.4 | |
| WUDI-MergingBackbone=ViT-L-142025.08 | 57.9 | 60.8 | |
| WUDI-MergingBackbone=ViT-B-322025.08 | 49.2 | 52.6 | |
| Pre-trainedBackbone=ViT-B-322025.08 | 47.3 | — |