Access to the multi-organ dataset: Download the dataset from the link. Convert them to numpy format, clip the images within [-125, 275], normalize each 3D image to [0, 1], and extract 2D slices from 3D volume for training cases while keep the 3D volume in h5 format for testing cases.