Class-Incremental Learning on VTAB-Sim50
1.05Final Average ForgettingRanPAC
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| RanPACScenario=Varying, Backbone=ViT-B/16, Pre-trained=ImageNet-21k fine-tuned on ImageNet-1K2026.01 | 1.05 | 1 | 2.5 | |
| PROTEUSScenario=Varying, Backbone=ViT-B/16, Pre-trained=ImageNet-21k fine-tuned on ImageNet-1K, LoRA rank r=42026.01 | 1.41 | 2 | 2.17 | |
| SPromptScenario=Varying, Backbone=ViT-B/16, Pre-trained=ImageNet-21k fine-tuned on ImageNet-1K2026.01 | 3.86 | 3 | 2.67 | |
| DualPromptScenario=Varying, Backbone=ViT-B/16, Pre-trained=ImageNet-21k fine-tuned on ImageNet-1K2026.01 | 8.74 | 4 | 4.83 | |
| CODAPromptScenario=Varying, Backbone=ViT-B/16, Pre-trained=ImageNet-21k fine-tuned on ImageNet-1K2026.01 | 9.67 | 5 | 3.17 | |
| L2PScenario=Varying, Backbone=ViT-B/16, Pre-trained=ImageNet-21k fine-tuned on ImageNet-1K2026.01 | 17.73 | 6 | 5.67 |