Long-Tail Class-Incremental Learning on ImageNet-R 20 tasks rho=0.01
70Accuracy (Last Task)Ours
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OursBackbone=ViT-B/162026.03 | 70 | 1.8 | |
| APARTBackbone=ViT-B/162026.03 | 63 | 2.5 | |
| MG-CLIPBackbone=ViT-B/162026.03 | 62.9 | 2.5 | |
| DAPBackbone=ViT-B/162026.03 | 61.9 | 2.4 | |
| RAPFBackbone=ViT-B/162026.03 | 61.3 | 2.4 | |
| GMMBackbone=ViT-B/162026.03 | 60.7 | 2.2 | |
| CODAPromptBackbone=ViT-B/162026.03 | 52.2 | 4.6 | |
| ISPCBackbone=ViT-B/162026.03 | 51.8 | 5.3 | |
| PODNET + GVAlignBackbone=ViT-B/162026.03 | 50.8 | 5.9 | |
| PODNET + LWSBackbone=ViT-B/162026.03 | 47.6 | 6.1 | |
| L2PBackbone=ViT-B/162026.03 | 47.6 | 5.9 | |
| DualPromptBackbone=ViT-B/162026.03 | 47.2 | 5.6 | |
| PriViLegeBackbone=ViT-B/162026.03 | 45.9 | 3 | |
| LFM+MMSBackbone=ViT-B/162026.03 | 15.7 | 28.9 |