Long-Tail Class-Incremental Learning on ImageNet-R 10 tasks (rho=0.01)
72.1Accuracy Last TaskOurs
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OursBackbone=ViT-B/162026.03 | 72.1 | 1.9 | |
| APARTBackbone=ViT-B/162026.03 | 65.2 | 1.9 | |
| MG-CLIPBackbone=ViT-B/162026.03 | 64.9 | 2 | |
| DAPBackbone=ViT-B/162026.03 | 64.8 | 2.1 | |
| RAPFBackbone=ViT-B/162026.03 | 63.5 | 2.1 | |
| GMMBackbone=ViT-B/162026.03 | 62.3 | 2.3 | |
| CODAPromptBackbone=ViT-B/162026.03 | 56.8 | 4.5 | |
| ISPCBackbone=ViT-B/162026.03 | 52.1 | 5.5 | |
| PODNET + GVAlignBackbone=ViT-B/162026.03 | 51.3 | 6.2 | |
| DualPromptBackbone=ViT-B/162026.03 | 51.1 | 5 | |
| L2PBackbone=ViT-B/162026.03 | 50.9 | 5.4 | |
| PriViLegeBackbone=ViT-B/162026.03 | 50.4 | 3.3 | |
| PODNET + LWSBackbone=ViT-B/162026.03 | 48.3 | 6.5 | |
| LFM+MMSBackbone=ViT-B/162026.03 | 16.8 | 27.1 |