I'm Simone Alghisi, a student from the University of Trento. Together with a couple of colleagues, we noticed that there are mainly two problems in the proposed code:
firstly, there is no Classifier.py under the libs folder, meaning that it is not possible to train supervisedly a model;
secondly, the dataset Indigo Mobile has images of 330x330. However, the proposed U-Net implementation accepts images with a size compatible with a power of two, meaning that it is impossible to train a one-class SVM.
First of all, I would like to thanks you for your interest to our work.
The Classifier.py has been added.
To satisfy the U-Net requirement, you can use the central cropping of the appropriate size assuring that the image starts from the fool (non-cropped) symbol at top left.
I'm Simone Alghisi, a student from the University of Trento. Together with a couple of colleagues, we noticed that there are mainly two problems in the proposed code:
Sincerely.