This code implements the recovery of image x from the undersampled measurements y when the pair (x,y) is not avaialble for training. However, we have a small amount of ground-truth x from y. This is important for medical imaging applications where usually one doesn't have access to high-resolution datastes for all organs.
python3 wgancs_main.py
--dataset_train ./Knee-highresolution-19cases/train_small
--dataset_label ./Knee-highresolution-19cases/partial_labels
--dataset_test ./Knee-highresolution-19cases/test_small
--sampling_pattern ./Knee-highresolution-19cases/sampling_pattern/mask_5fold_160_128_knee_vdrad.mat
--sample_size 320 --sample_size_y 256
--batch_size 8 --sample_test 24
--summary_period 1700
--train_dir ./train_dir/exp29
--checkpoint_dir ./checkpoint/exp29
--wgan_gp True
--activation lrelu
--learning_rate_start 5e-5
For medical image reconstruction we adopt the MRI datasets available at the https://www.mridata.org made available as a result of a joint collaboration between Stanford & UC Berkeley. It includes a 20 3D Knee images that have a high resoltuion of 192x320x256. 192 2D axial slices are collected from all patients to form the training and test datasets.
-- The input files have .jpg format in the train and test folders
-- The sampling mask is randomly generated based on a avariable density with radial view ordering sampling technique. The Matlab code is avialble at http://mrsrl.stanford.edu/~jycheng/software.html