This repository accompanies the paper Data-Centric Diet: Effective Dataset Pruning for Medical Image Segmentation and contains the basic code for replicating the training and computations in it. We found easy and hard to learn example for deep learning model by DAD(Dynamic average dice) Score.
The main requirements are pytorch 1.4.0 with python 3.9.1.
To train one independent run of 3D-UNet on MSD-Pancreas (the full dataset), first set up params args.dataset='msd' in params_msd_only, then from <ROOT> execute
python run/train_vnet.py This is a demo presentation that may encounter bugs or issues that may not run properly. The complete code will be released after the paper is received.