Fraction of images that do not change the class label depending on the choice of regularization parameter . For this analysis, 40 images were chosen from each of the five classes, such that there were
Version 1Fraction of images that do not change the class label depending on the choice of regularization parameter . For this analysis, 40 images were chosen from each of the five classes, such that there were 80 images for the “healthy to DR” direction and 120 for “DR to healthy”. For 73/80 and 100/120, class labels were correctly predicted for the original image. Then we evaluated the class label of the corresponding counterfactual.
No license available
Fraction of images that do not change the class label depending on the choice of regularization parameter . For this analysis, 40 images were chosen from each of the five classes, such that there were 80 images for the “healthy to DR” direction and 120 for “DR to healthy”. For 73/80 and 100/120, class labels were correctly predicted for the original image. Then we evaluated the class label of the corresponding counterfactual.