Table 8 Domain generalization performance: evaluation of architectures (discovered on source datasets) on unseen external target datasets without retraining
Source dataset | Target external dataset | Backbone | Method | Performance (%)↑ |
|---|---|---|---|---|
BreakHis | SkinTumor27 | ShuffleNet | Random Search | 39.21 ± 0.57 (Prec@1) |
BreakHis | SkinTumor27 | ShuffleNet | Pathology-NAS | 74.50 ± 0.98 (Prec@1) |
BreakHis | SkinTumor27 | ViT | Random Search | 73.52 ± 0.94 (Prec@1) |
BreakHis | SkinTumor27 | ViT | Pathology-NAS | 82.35 ± 0.29 (Prec@1) |
BreakHis | SkinTumor27 | MobileNetV3 | Random Search | 45.30 ± 0.70 (Prec@1) |
BreakHis | SkinTumor27 | MobileNetV3 | Pathology-NAS | 78.10 ± 0.60 (Prec@1) |
PanNuke | Polyp28 | U-Net | Random Search | 39.18 ± 0.27 (Dice) |
PanNuke | Polyp28 | U-Net | Pathology-NAS | 62.07 ± 0.45 (Dice) |
