Table 2 Best results on the validation set for each encoder and training strategy. The best performing combination is shown in bold
Scenario | Encoder name | Best of | Batch size | Step | Learning rate | DS \(\uparrow\) |
|---|---|---|---|---|---|---|
1 | ResNet-18 | 60 | 128 | 4900 | \(1.43 \cdot 10^{-4}\) | 0.8732 |
1 | ResNet-34 | 50 | 384 | 4280 | \(2.37 \cdot 10^{-4}\) | 0.8655 |
1 | ResNet-50 | 30 | 256 | 3540 | \(3.97 \cdot 10^{-4}\) | 0.8982 |
1 | ResNet-101 | 20 | 256 | 4020 | \(5.40 \cdot 10^{-4}\) | 0.8995 |
2 | ResNet-18 | 60 | 128 | 2160 | \(5.40 \cdot 10^{-4}\) | 0.8862 |
2 | ResNet-34 | 50 | 384 | 1680 | \(5.11 \cdot 10^{-4}\) | 0.8837 |
2 | ResNet-50 | 30 | 128 | 2960 | \({\textbf {5.40}} \cdot {\textbf {10}}^{{\textbf {-4}}}\) | 0.9048 |
2 | ResNet-101 | 20 | 256 | 440 | \(7.35 \cdot 10^{-4}\) | 0.9011 |