matlab_to_numpy.py

"""
Read all the Matlab files in the 'data' directory and export 3 numpy arrays:
- labels.npy
- images.npy
- masks.npy

Usage: python matlab_to_numpy.py ~/brain_tumor_dataset
"""

import os
import argparse
import sys
import numpy as np
import hdf5storage
import cv2


class NoDataFound(Exception):
    pass


def dir_path(path):
    """Check the path and the existence of a data directory"""
    # replace '\' in path for Windows users
    path = path.replace('\\', '/')
    data_path = os.path.join(path, 'data').replace('\\', '/')

    if os.path.isdir(data_path):
        return path
    elif os.path.isdir(path):
        raise NoDataFound('Could not find a "data" folder inside directory. {} does not exist.'
                          .format(data_path))
    else:
        raise NotADirectoryError(path)


parser = argparse.ArgumentParser()
parser.add_argument('path', help='path to the brain_tumor_dataset directory', type=dir_path)
parser.add_argument('--image-dimension', '-d', default=512, help='dimension of the image', type=int)
args = parser.parse_args()

labels = []
images = []
masks = []

data_dir = os.path.join(args.path, 'data').replace('\\', '/')
files = os.listdir(data_dir)
for i, file in enumerate(files, start=1):
    if i % 10 == 0:
        # print the percentage of images loaded
        sys.stdout.write('\r[{}/{}] images loaded: {:.1f} %'
                         .format(i, len(files), i / float(len(files)) * 100))
        sys.stdout.flush()

    # load matlab file with hdf5storage as scipy.io.loadmat does not support v7.3 files
    mat_file = hdf5storage.loadmat(os.path.join(data_dir, file))['cjdata'][0]

    # resize image and mask to a unique size
    image = cv2.resize(mat_file[2], dsize=(args.image_dimension, args.image_dimension),
                       interpolation=cv2.INTER_CUBIC)
    mask = cv2.resize(mat_file[4].astype('uint8'), dsize=(args.image_dimension, args.image_dimension),
                      interpolation=cv2.INTER_CUBIC)

    labels.append(int(mat_file[0]))
    images.append(image)
    masks.append(mask.astype(bool))

sys.stdout.write('\r[{}/{}] images loaded: {:.1f} %'
                 .format(i, len(files), i / float(len(files)) * 100))
sys.stdout.flush()

labels = np.array(labels)
images = np.array(images)
masks = np.array(masks)

print('\nlabels:', labels.shape)
print('images:', images.shape)
print('masks:', masks.shape)

np.save(os.path.join(args.path, 'labels.npy'), labels)
np.save(os.path.join(args.path, 'images.npy'), images)
np.save(os.path.join(args.path, 'masks.npy'), masks)

print('labels.npy, images.npy, masks.npy saved in', args.path)