def read_data(idx):
name, label = sample_list[idx]
d = np.load(os.path.join(DATA_DIR, name + '.npz'))
pix = zero_center(normalize(d['pix_resampled']))
# set all outside regions to zero
pix *= d['segmented_lungs_fill_dilated']
return pix, label
# Load training labels
sample_list = []
with open(LABLE_FILE_TRAIN) as f:
for name, label in csv.reader(f):
sample_list.append((name, float(label)))
num_samples = len(sample_list)
print('number of samples:', num_samples)
# Pre-load all the data
data_list = [None] * num_samples
label_list = [None] * num_samples
for idx in range(num_samples):
print('pre-loading data %d / %d' % (idx + 1, len(sample_list)))
name, label = sample_list[idx]
d = np.load(os.path.join(DATA_DIR, name + '.npz'))
pix = zero_center(normalize(d['pix_resampled_add0'])).astype(np.float32)
data_list[idx] = pix
label_list[idx] = label
d.close()
batch_data = np.zeros((N, Z, Y, X, 1), np.float32)
batch_labels = np.zeros((N, 1), np.float32)
def get_random_samples():
inds = np.random.choice(num_samples, N, replace=False)
for i, idx in enumerate(inds):
batch_data[i, ..., 0] = data_list[idx]
batch_labels[i] = label_list[idx]
return batch_data, batch_labels
def read_data(idx):
name, label = sample_list[idx]
d = np.load(os.path.join(DATA_DIR, name + '.npz'))
pix = zero_center(normalize(d['pix_resampled_add0']))
return pix, label