Geekiter / IRA-Unet

IRA-Unet: Inception Residual Attention Unet in Adversarial Network for Cardiac MRI Segmentation

IRA-Unet

Introduction

This repository contains the implementation for automated cardiac segmentation introduced in the following paper: "IRA-Unet: Inception Residual Attention Unet in Adversarial Network for Cardiac MRI Segmentation"

Steps to train and test the model:

1.Register and download ACDC-2017 dataset from https://www.creatis.insa-lyon.fr/Challenge/acdc/index.html

2.Run the script preprocess.py.

python preprocess.py --data-root your DATA_DIR

3.A folder named loc192 will be created which contain preprocessed and croped train and validation dataset.

4.Run the script main.py.

python main.py --data-root your DATA_DIR --save-path your OUT_DIR

The segmented image of test set will be saved in outputs

Steps to test the pretrained model:

1.To reproduce the results, download weights of our best model from here

2.Put the last.ckpt file in ckpt folder

2.Run the script predict.py.

python predict.py --data-root your DATA_DIR --save-path your OUT_DIR

Requirements

The code is tested on Ubuntu 20.04 with the following components:

Software Python 3.8 pytorch 1.13 CUDA 11.8

Logs

To launch the tensorboard instance run

tensorboard --logdir 'logs/IRA-Unet'

It will give a view on the evolution of the loss for both the training and validation data.