Training Time and Accuracy on VTAB (Continual Learning)
1.7Average Training Time per TaskFly-CL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Fly-CLBackbone=ViT-B/16, Task Sequence=Longer2025.10 | 1.7 | 0.29 | 96.54 | |
| F-OALBackbone=ViT-B/16, Task Sequence=Longer2025.10 | 2.17 | 0.71 | 94.96 | |
| Fly-CLBackbone=ViT-B/16, Pre-trained=true2025.10 | 2.48 | 0.34 | 96.54 | |
| Fly-CLBackbone=ResNet-502025.10 | 2.53 | 0.34 | 94 | |
| F-OALBackbone=ResNet-502025.10 | 2.76 | 0.55 | 26.15 | |
| F-OALBackbone=ViT-B/16, Pre-trained=true2025.10 | 3.19 | 1.03 | 94.68 | |
| MoE-AdapterBackbone=ViT-B/16, Task Sequence=Longer2025.10 | 29.25 | 26.81 | 83.24 | |
| DualpromptBackbone=ViT-B/16, Task Sequence=Longer2025.10 | 29.44 | 26.62 | 78.46 | |
| InfLoRABackbone=ViT-B/16, Task Sequence=Longer2025.10 | 29.71 | 26.92 | 81.09 | |
| SEMABackbone=ViT-B/16, Task Sequence=Longer2025.10 | 30.36 | 27.55 | 84.56 | |
| L2PBackbone=ViT-B/16, Task Sequence=Longer2025.10 | 32.17 | 29.31 | 71.84 | |
| MoE-AdapterBackbone=ViT-B/16, Pre-trained=true2025.10 | 32.28 | 26.06 | 86.39 | |
| InfLoRABackbone=ViT-B/16, Pre-trained=true2025.10 | 35.8 | 29.26 | 88.73 | |
| DualPromptBackbone=ViT-B/16, Pre-trained=true2025.10 | 38.14 | 31.64 | 80.85 | |
| RanPACBackbone=ViT-B/16, Task Sequence=Longer2025.10 | 41.49 | 40.02 | 93.62 | |
| L2PBackbone=ViT-B/16, Pre-trained=true2025.10 | 42.1 | 36.45 | 81.24 | |
| SEMABackbone=ViT-B/16, Pre-trained=true2025.10 | 45.5 | 39.82 | 91.18 | |
| RanPACBackbone=ResNet-502025.10 | 50.15 | 47.94 | 92.8 | |
| RanPACBackbone=ViT-B/16, Pre-trained=true2025.10 | 63.68 | 61.44 | 94.16 | |
| EASEBackbone=ViT-B/16, Task Sequence=Longer2025.10 | 64.68 | 57.24 | 90.26 | |
| EASEBackbone=ViT-B/16, Pre-trained=true2025.10 | 108.35 | 94.73 | 94.02 |