Fig. 3
From: On the role of deep learning model complexity in adversarial robustness for medical images
Saliency maps of standard trained CBR-LargeT (a, c, e) and Resnet50 (b, d, & f) for Chest X-ray, Dermoscopy & OCT datasets. Unperturbed images with predicted labels and corresponding saliency maps were visualized on row 1 (no attack) for (a–f), respectively. Imperceptibly perturbed images with predicted labels and corresponding saliency maps were visualized on row 2 (PGD attack) for (a–f), respectively. Saliency maps of CBR-LargeT are more concentrated on the regions of interest whereas Resnet50 includes attention regions that are more sporadic on areas that do not contribute to the classification of the disease