Domain Adaptation on ImageNet (Source: 1K, Targets: Sketch, V2, A, R) (test)
0.79Accuracy on Source Domain (1K)LoRAmul + VPTadd + PACE
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LoRAmul + VPTadd + PACEBackbone=ViT-B/16, Pre-trained=ImageNet-21K2024.09 | 0.79 | 0.318 | 0.694 | 0.163 | 0.352 | 0.463 | |
| GLORABackbone=ViT-B/16, Pre-trained=ImageNet-21K2024.09 | 0.783 | 0.306 | 0.675 | 0.133 | 0.31 | 0.441 | |
| LoRAmul + VPTaddBackbone=ViT-B/16, Pre-trained=ImageNet-21K2024.09 | 0.783 | 0.306 | 0.685 | 0.141 | 0.325 | 0.448 | |
| NOAHBackbone=ViT-B/16, Pre-trained=ImageNet-21K2024.09 | 0.715 | 0.248 | 0.661 | 0.119 | 0.285 | 0.405 | |
| LoRABackbone=ViT-B/16, Pre-trained=ImageNet-21K2024.09 | 0.708 | 0.2 | 0.593 | 0.069 | 0.233 | 0.36 | |
| AdapterBackbone=ViT-B/16, Pre-trained=ImageNet-21K2024.09 | 0.705 | 0.164 | 0.591 | 0.055 | 0.221 | 0.347 | |
| VPTBackbone=ViT-B/16, Pre-trained=ImageNet-21K2024.09 | 0.705 | 0.183 | 0.58 | 0.046 | 0.232 | 0.347 | |
| LinearBackbone=ViT-B/16, Pre-trained=ImageNet-21K2024.09 | 0.679 | 0.144 | 0.608 | 0.094 | 0.256 | 0.356 | |
| FullBackbone=ViT-B/16, Pre-trained=ImageNet-21K2024.09 | 0.639 | 0.185 | 0.525 | 0.032 | 0.212 | 0.318 |