Cross-dataset Continual Learning on CIFAR100-I2C Transfer from ImageNet100 (test)
82.3AccuracyAttriCLIP
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| AttriCLIPMemory=02023.05 | 82.3 | 0.9 | — | |
| DualPromptMemory=0, Expansion Scheme=Method-12023.05 | 80.7 | -5.8 | — | |
| DualPromptMemory=0, Expansion Scheme=Method-22023.05 | 74.7 | -9.4 | — | |
| ARIMemory=2000, Expansion Scheme=Method-12023.05 | 74.5 | -6.4 | — | |
| Continual-CLIPMemory=02023.05 | 66.7 | 0 | — | |
| CoOpMemory=1000, Expansion Scheme=Method-12023.05 | 61.1 | -6.5 | — | |
| ARIMemory=2000, Expansion Scheme=Method-22023.05 | 59.9 | -19.8 | — | |
| CoOpMemory=1000, Expansion Scheme=Method-22023.05 | 59 | -8.6 | — | |
| iCaRLMemory=2000, Expansion Scheme=Method-12023.05 | 49.7 | 0.2 | — | |
| iCaRLMemory=2000, Expansion Scheme=Method-22023.05 | 46.5 | -2.6 | — | |
| AttriCLIPBackbone=ViT-L/14, Batch size=322024.03 | — | — | 78.3 | |
| AttriCLIP + CLAP (Ours)Backbone=ViT-L/14, Batch size=322024.03 | — | — | 84.56 | |
| CLAP (Ours)Backbone=ViT-L/14, Batch size=322024.03 | — | — | 83.83 | |
| Continual-CLIPBackbone=ViT-L/14, Batch size=322024.03 | — | — | 54.9 | |
| CoOpBackbone=ViT-L/14, Batch size=322024.03 | — | — | 55.4 | |
| CoOp + CLAP (Ours)Backbone=ViT-L/14, Batch size=322024.03 | — | — | 82.63 | |
| DualPromptBackbone=ViT-L/14, Batch size=322024.03 | — | — | 67.1 | |
| MaPLeBackbone=ViT-L/14, Batch size=322024.03 | — | — | 76.2 | |
| MaPLe + CLAP (Ours)Backbone=ViT-L/14, Batch size=322024.03 | — | — | 83.6 | |
| PROOFBackbone=ViT-L/14, Batch size=322024.03 | — | — | 82.59 |