Table 8 Impact of different data augmentation strategies on domain adaptation performance across datasets and backbones.
Gaussian blur | Erasing | Grayscale | ImageCLEF | Office-Home (ResNet-50) | Office-Home (ViT) | VisDA-2017 (ResNet-101) |
|---|---|---|---|---|---|---|
\(\checkmark\) | Â | Â | 91.03 | 70.81 | 86.37 | 87.80 |
| Â | \(\checkmark\) | Â | 90.37 | 71.40 | 86.38 | 87.20 |
| Â | Â | \(\checkmark\) | 90.36 | 71.44 | 86.35 | 86.25 |
\(\checkmark\) | \(\checkmark\) | Â | 90.75 | 71.57 | 86.46 | 84.53 |
\(\checkmark\) | Â | \(\checkmark\) | 90.19 | 71.95 | 86.49 | 84.68 |
| Â | \(\checkmark\) | \(\checkmark\) | 89.84 | 72.29 | 86.63 | 84.19 |
\(\checkmark\) | \(\checkmark\) | \(\checkmark\) | 90.53 | 73.86 | 87.21 | 87.01 |
