Dual-ArbNet-PyTorch

Official Pytorch implementation of "Dual Arbitrary Scale Super-Resolution for Multi-Contrast MRI" (MICCAI2023)

Paper

Requirements

• Python 3.9
• asposestorage==1.0.2
• imageio==2.22.4
• matplotlib==3.6.2
• numpy==1.23.5
• opencv_python==4.6.0.66
• scikit_image==0.19.3
• scipy==1.10.1
• skimage==0.0
• thop==0.1.1.post2209072238
• torch==1.13.0
• torchvision==0.14.0
• tqdm==4.64.1

Train

1. Prepare training data

Downkload fastMRI dataset and IXI dataset.

Filter the multi contrast MRI datasets.

2. Begin to train

Run ./main.sh to train on the training dataset. Please update name_train, dir_data, save, ref_mat, ref_list in the bash file as your needs.

Quick Test on An LR MR Image

Download pre-trained weights and put it in the experiment folder.

Run ./test_save.sh to enlarge an LR image to an arbitrary size. Please update dir_data and pre_train in the bash file as your_path.

Visual Results

SR with Arbitrary Scale Factors

You can change the --scale ./test_save.sh to obtain the results of different scale factors. You can also change the --ref_type_test ./test_save.sh to use HR(1) or LR(2) reference image.