Continual Learning Performance Metrics (A10, Avg.) on ImageNet-A (test)
65.01A10Joint Train
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Joint TrainBackbone=ViT-B/16, Pre-trained=ImageNet-21K, Training Strategy=Joint2026.02 | 65.01 | 74.1 | |
| EWC-LoRABackbone=ViT-B/16, Pre-trained=ImageNet-21K, Training Strategy=Sequential2026.02 | 59.89 | 68.33 | |
| CL-LoRABackbone=ViT-B/16, Pre-trained=ImageNet-21K, Training Strategy=Sequential2026.02 | 57.62 | 70.76 | |
| SD-LoRABackbone=ViT-B/16, Pre-trained=ImageNet-21K, Training Strategy=Sequential2026.02 | 55.23 | 66.1 | |
| BiLoRABackbone=ViT-B/16, Pre-trained=ImageNet-21K, Training Strategy=Sequential2026.02 | 51.05 | 62.82 | |
| InfLoRABackbone=ViT-B/16, Pre-trained=ImageNet-21K, Training Strategy=Sequential2026.02 | 50.75 | 64.36 | |
| DualPromptBackbone=ViT-B/16, Pre-trained=ImageNet-21K, Training Strategy=Sequential2026.02 | 45.49 | 54.68 | |
| CODA-PromptBackbone=ViT-B/16, Pre-trained=ImageNet-21K, Training Strategy=Sequential2026.02 | 45.36 | 57.03 | |
| L2PBackbone=ViT-B/16, Pre-trained=ImageNet-21K, Training Strategy=Sequential2026.02 | 42.94 | 51.4 | |
| Vanilla LoRABackbone=ViT-B/16, Pre-trained=ImageNet-21K, Training Strategy=Sequential2026.02 | 40.01 | 58.28 | |
| FinetuneBackbone=ViT-B/16, Pre-trained=ImageNet-21K, Training Strategy=Sequential2026.02 | 32.85 | 54.55 |