Continual Learning on CUB
-0.037Backward Transfer (BwT)CLAP4CLIP
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CLAP4CLIPbase_method=AttriCLIP2024.03 | -0.037 | — | — | — | |
| CLAP4CLIPbase_method=MaPLe2024.03 | -0.102 | — | — | — | |
| CLAP4CLIPbase_method=CoOp2024.03 | -0.106 | — | — | — | |
| CLAP4CLIPbase_method=Continual-CLIP2024.03 | -0.117 | — | — | — | |
| Continual-CLIP2024.03 | -0.124 | — | — | — | |
| MaPLe2024.03 | -0.145 | — | — | — | |
| AttriCLIP2024.03 | -0.151 | — | — | — | |
| CoOp2024.03 | -0.162 | — | — | — | |
| Fly-CLBackbone=ViT-B/162025.10 | -3.8 | 89.97 | — | — | |
| RanPACBackbone=ViT-B/162025.10 | -4.16 | 88.95 | — | — | |
| F-OALBackbone=ViT-B/162025.10 | -5.43 | 85.61 | — | — | |
| EASEBackbone=ViT-B/162025.10 | -5.68 | 85.1 | — | — | |
| SEMABackbone=ViT-B/162025.10 | -9.94 | 73.66 | — | — | |
| InfLoRABackbone=ViT-B/162025.10 | -11.53 | 69.45 | — | — | |
| MoE-AdapterBackbone=ViT-B/162025.10 | -12.11 | 68.11 | — | — | |
| DualPromptBackbone=ViT-B/162025.10 | -12.12 | 67.9 | — | — | |
| L2PBackbone=ViT-B/162025.10 | -13.14 | 65.41 | — | — | |
| ACMapNumber of tasks (T)=20, Setting=Long-tailed step-imbalance2026.03 | — | — | 79.47 | 85.9 | |
| Aper-adapterNumber of tasks (T)=20, Setting=Long-tailed step-imbalance2026.03 | — | — | 79.3 | 85.78 | |
| CL-LoRANumber of tasks (T)=20, Setting=Long-tailed step-imbalance2026.03 | — | — | 67.9 | 81.41 | |
| EASENumber of tasks (T)=20, Setting=Long-tailed step-imbalance2026.03 | — | — | 79.4 | 86.59 | |
| InfLoRANumber of tasks (T)=20, Setting=Long-tailed step-imbalance2026.03 | — | — | 49.32 | 62.52 | |
| JointBackbone=ViT-B16, Pre-training=ImageNet21K, Evaluation Protocol=Offline, Initial Classes=2, New Classes per Task=12024.02 | — | 91.1 | — | — | |
| One-ANumber of tasks (T)=20, Setting=Long-tailed step-imbalance2026.03 | — | — | 81.1 | 87.58 | |
| RanDumbBackbone=ViT-B16, Pre-training=ImageNet21K, Evaluation Protocol=Offline, Initial Classes=2, New Classes per Task=12024.02 | — | 88.5 | — | — | |
| RanPAC-impBackbone=ViT-B16, Pre-training=ImageNet21K, Evaluation Protocol=Offline, Initial Classes=2, New Classes per Task=12024.02 | — | 89.6 | — | — | |
| SEMANumber of tasks (T)=20, Setting=Long-tailed step-imbalance2026.03 | — | — | 49.06 | 62 | |
| SLCABackbone=ViT-B16, Pre-training=ImageNet21K, Evaluation Protocol=Offline, Initial Classes=2, New Classes per Task=12024.02 | — | 82.1 | — | — |