Long-Tail Class-Incremental Learning on ImageNet-R 20 tasks rho=0.1
73.7Accuracy (Last Task)Ours
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OursBackbone=ViT-B/162026.03 | 73.7 | 1.2 | |
| APARTBackbone=ViT-B/162026.03 | 70.3 | 1.7 | |
| DAPBackbone=ViT-B/162026.03 | 70 | 2 | |
| MG-CLIPBackbone=ViT-B/162026.03 | 68.2 | 1.6 | |
| CODAPromptBackbone=ViT-B/162026.03 | 66.5 | 2.6 | |
| RAPFBackbone=ViT-B/162026.03 | 65.8 | 2 | |
| GMMBackbone=ViT-B/162026.03 | 65.7 | 2.1 | |
| DualPromptBackbone=ViT-B/162026.03 | 61.2 | 3.1 | |
| L2PBackbone=ViT-B/162026.03 | 60.1 | 3.3 | |
| PriViLegeBackbone=ViT-B/162026.03 | 57.8 | 3.1 | |
| ISPCBackbone=ViT-B/162026.03 | 55.9 | 5.9 | |
| PODNET + GVAlignBackbone=ViT-B/162026.03 | 52.9 | 5.8 | |
| PODNET + LWSBackbone=ViT-B/162026.03 | 49.5 | 6 | |
| LFM+MMSBackbone=ViT-B/162026.03 | 22.7 | 29.1 |