Deep Prototypical-Parts Ease Morphological Kidney Stone Identification and are Competitively Robust to Photometric Perturbations
This repository contains Python code and bash files to train and explain prototipical parts learned by the Prototypical Part Network (ProtoPNet).
This is work implements ProtoPNet and Prototipical descriptors to explain an image classification, quantify the influence of color hue, shape, texture, contrast, saturation and brithness in each prototype locally and globally and evaluate performance of the trained model under such perturbations.
Corresponding paper on ArXiv: Deep Prototypical-Parts Ease Morphological Kidney Stone Identification and are Competitively Robust to Photometric Perturbations
Some evaluation and visualization code used to obtain the paper results will be uploaded in the future.