-
Notifications
You must be signed in to change notification settings - Fork 0
/
createAug.py
247 lines (222 loc) · 13.6 KB
/
createAug.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
import sys
import time
import numpy as np
import os
from PIL import Image
import random
import torchio as tio
import nibabel as nib
import torch.nn.functional as nnf
import torch
import matplotlib.pyplot as plt
from tqdm import tqdm
import skimage.transform as skTrans
from data.dataLoader_dataset import save3Dimage_numpy
from util.util import save3Dimage_numpy
import scipy
import nibabel as nib
def make_dataset(dir, max_dataset_size=float("inf")):
images = []
assert os.path.isdir(dir), '%s is not a valid directory' % dir
for root, _, fnames in sorted(os.walk(dir)):
for fname in fnames:
# print(fname, fname.endswith("nii.gz"))
if fname.endswith(".nii.gz") or fname.endswith(".npz") or fname.endswith(".nii"):
path = os.path.join(root, fname)
images.append(path)
return images[:min(max_dataset_size, len(images))]
def get_transform(params=None, convert=False, no_aug=True):
transform_list = []
# Augmentations:
if not no_aug:
print("Using Spatial Aug")
transform_list += [tio.RandomAnisotropy()]
max_displacement = 15, 15, 0
spatial_transforms = {
tio.RandomElasticDeformation(max_displacement=5, num_control_points=7, locked_borders=2) : 0.75
,tio.RandomAffine(scales=(0.9, 1.2), degrees=10, isotropic=True, image_interpolation='nearest'): 0.25
}
transform_list += [tio.OneOf(spatial_transforms)]
transform_list += [tio.RandomElasticDeformation(max_displacement=max_displacement, locked_borders=2)]
if convert:
print("Rescale Intensity")
transform_list += [tio.RescaleIntensity((-1, 1))]
return tio.Compose(transform_list)
class DataLoaderDataset():
def __init__(self, dataroot, no_aug=False, test=False, only_resize=False):
"""Initialize this dataset class.
Parameters:
opt (Option class) -- stores all the experiment flags; needs to be a subclass of BaseOptions
"""
self.no_aug = no_aug
if test:
print("Test data")
self.dir_ct = os.path.join(dataroot, 'testct') # create a path '/path/to/data/ct'
self.dir_mr = os.path.join(dataroot, 'testmr') # create a path '/path/to/data/mr'
self.dir_ct_label = os.path.join(dataroot, 'testct_labels') # create a path '/path/to/data/ct'
self.dir_mr_label = os.path.join(dataroot, 'testmr_labels') # create a path '/path/to/data/mr'
else:
self.dir_ct = os.path.join(dataroot, 'trainct') # create a path '/path/to/data/ct'
self.dir_mr = os.path.join(dataroot, 'trainmr') # create a path '/path/to/data/mr'
self.dir_ct_label = os.path.join(dataroot, 'trainct_labels') # create a path '/path/to/data/ct'
self.dir_mr_label = os.path.join(dataroot, 'trainmr_labels') # create a path '/path/to/data/mr'
self.ct_paths = sorted(make_dataset(self.dir_ct, float("inf"))) # load images from '/path/to/data/ct'
self.mr_paths = sorted(make_dataset(self.dir_mr, float("inf"))) # load images from '/path/to/data/mr'
self.ct_paths_label = sorted(make_dataset(self.dir_ct_label, float("inf"))) # load images from '/path/to/data/ct'
self.mr_paths_label = sorted(make_dataset(self.dir_mr_label, float("inf"))) # load images from '/path/to/data/mr'
self.ct_size = len(self.ct_paths) # get the size of dataset ct
self.mr_size = len(self.mr_paths) # get the size of dataset mr
self.ct_size_label = len(self.ct_paths_label) # get the size of dataset ct
self.mr_size_label = len(self.mr_paths_label) # get the size of dataset mr
mrtoct = 'mrtoct'
if test or only_resize:
self.transform_ct = get_transform(no_aug=True, convert=True)
self.transform_mr = get_transform(no_aug=True, convert=True)
else:
self.transform_ct = get_transform(no_aug=False, convert=True)
self.transform_mr = get_transform(no_aug=False, convert=True)
self.gpu_ids = ["0", "1"]
self.device = torch.device('cuda:{}'.format(self.gpu_ids[0])) if self.gpu_ids else torch.device('cpu') # get device name: CPU or GPU
self.only_resize = only_resize
self.test = test
self.aug = True
def __getitem__(self, index):
"""Return a data point and its metadata information.
Parameters:
index (int) -- a random integer for data indexing
Returns a dictionary that contains ct, mr, ct_paths and mr_paths
ct (tensor) -- an image in the input domain
mr (tensor) -- its corresponding image in the target domain
ct_paths (str) -- image paths
mr_paths (str) -- image paths
"""
ct_path = self.ct_paths[index % self.ct_size] # make sure index is within then range
ct_path_label = self.ct_paths_label[index % self.ct_size] # make sure index is within then range
# index_mr = random.randint(0, self.mr_size - 1)
index_mr = index
mr_path = self.mr_paths[index_mr]
mr_path_label = self.mr_paths_label[index_mr]
if ".npz" in ct_path:
ct_label = np.load(ct_path_label)['arr_0']
ct_img = np.load(ct_path)['arr_0']
mr_label = np.load(mr_path_label)['arr_0']
mr_img = np.load(mr_path)['arr_0']
else:
mr_img = nib.load(mr_path)
mr_img = nib.as_closest_canonical(mr_img)
mr_img = np.expand_dims(np.array(mr_img.dataobj), axis=0)
mr_label = nib.load(mr_path_label)
mr_label = nib.as_closest_canonical(mr_label)
mr_label = np.expand_dims(np.array(mr_label.dataobj), axis=0)
ct_img = nib.load(ct_path)
ct_img = nib.as_closest_canonical(ct_img)
ct_img = np.expand_dims(np.array(ct_img.dataobj), axis=0)
ct_label = nib.load(ct_path_label)
ct_label = nib.as_closest_canonical(ct_label)
ct_label = np.expand_dims(np.array(ct_label.dataobj), axis=0)
numpy_reader = lambda path: np.load(path)['arr_0'], np.eye(4)
ct_subject = tio.Subject(
t1 = tio.ScalarImage(tensor=ct_img),
label = tio.LabelMap(tensor=ct_label)
)
mr_subject = tio.Subject(
t1 = tio.ScalarImage(tensor=mr_img),
label = tio.LabelMap(tensor=mr_label)
)
# apply image transformation
# ------------------------------------------------
if self.aug:
if self.transform_ct is None:
ct_img = ct_subject.t1.data
mr_img = mr_subject.t1.data
ct_label = ct_subject.label.data
mr_label = mr_subject.label.data
# if index % self.ct_size == 0:
# save3Dimage_numpy(ct_label.squeeze(), "Images_Print/ct_{}_label".format(index))
else:
ct_subject_transformed = self.transform_ct(ct_subject)
mr_subject_transformed = self.transform_mr(mr_subject)
print("Aug {} done".format(index))
ct_img = ct_subject_transformed.t1.data.numpy()
mr_img = mr_subject_transformed.t1.data.numpy()
ct_label = ct_subject_transformed.label.data.numpy()
mr_label = mr_subject_transformed.label.data.numpy()
else:
orientation_list = [1] if self.test else [0,1,9,10,11]
if (index % self.mr_size) in orientation_list:
# if (index % self.mr_size) in [1]:
# print("Before", ct_img.shape, mr_img.shape)
mr_img = mr_img[:,112:400,:,112:400]
mr_label = mr_label[:,112:400,:,112:400]
mr_img = skTrans.resize(mr_img, (mr_img.shape[0], 256, 256, 256), order=1, preserve_range=True, anti_aliasing=True)
mr_label = skTrans.resize(mr_label, (mr_label.shape[0], 256, 256, 256), order=0, preserve_range=True, anti_aliasing=False)
ct_scale = ct_img.shape[1]/256
ct_img = skTrans.resize(ct_img, (ct_img.shape[0], int(ct_img.shape[1]/ct_scale), int(ct_img.shape[2]/ct_scale), int(ct_img.shape[3]/ct_scale)), order=1, preserve_range=True, anti_aliasing=True)
ct_label = skTrans.resize(ct_label, (ct_label.shape[0], int(ct_label.shape[1]/ct_scale), int(ct_label.shape[2]/ct_scale), int(ct_label.shape[3]/ct_scale)), order=0, preserve_range=True, anti_aliasing=False)
# print("After", ct_img.shape, mr_img.shape)
else:
# print("Before", ct_img.shape, mr_img.shape)
# mr_img = mr_img[:,:,mr_img.shape[2]//2-mr_img.shape[1]//2:mr_img.shape[2]//2+mr_img.shape[1]//2,:]
# mr_label = mr_label[:,:,mr_label.shape[2]//2-mr_label.shape[1]//2:mr_label.shape[2]//2+mr_label.shape[1]//2,:]
ct_scale = ct_img.shape[1]/256
mr_scale = mr_img.shape[1]/256
smaller_than_128 = mr_img.shape[1] < 128
if ct_scale != 1 or mr_scale!= 1:
ct_img = skTrans.resize(ct_img, (ct_img.shape[0], int(ct_img.shape[1]/ct_scale), int(ct_img.shape[2]/ct_scale), int(ct_img.shape[3]/ct_scale)), order=1, preserve_range=True, anti_aliasing=True)
ct_label = skTrans.resize(ct_label, (ct_label.shape[0], int(ct_label.shape[1]/ct_scale), int(ct_label.shape[2]/ct_scale), int(ct_label.shape[3]/ct_scale)), order=0, preserve_range=True, anti_aliasing=False)
if smaller_than_128:
mr_img = skTrans.resize(mr_img, (mr_img.shape[0], 128, mr_img.shape[2], mr_img.shape[3]), order=1, preserve_range=True, anti_aliasing=True)
mr_label = skTrans.resize(mr_label, (mr_label.shape[0], 128, mr_label.shape[2], mr_label.shape[3]), order=0, preserve_range=True, anti_aliasing=False)
# else:
# mr_img = skTrans.resize(mr_img, (mr_img.shape[0], int(mr_img.shape[1]/mr_scale), int(mr_img.shape[2]/mr_scale), int(mr_img.shape[3]/mr_scale)), order=1, preserve_range=True, anti_aliasing=True)
# mr_label = skTrans.resize(mr_label, (mr_label.shape[0], int(mr_label.shape[1]/mr_scale), int(mr_label.shape[2]/mr_scale), int(mr_label.shape[3]/mr_scale)), order=0, preserve_range=True, anti_aliasing=False)
# print("After", ct_img.shape, mr_img.shape)
# save3Dimage_numpy(ct_label.squeeze(), "Images_Print/ct_{}_label".format(index))
# save3Dimage_numpy(ct_img.squeeze(), "Images_Print/ct_{}.png".format(index))
# save3Dimage_numpy(mr_label.squeeze(), "Images_Print/mr_{}_label.png".format(index))
# save3Dimage_numpy(mr_img.squeeze(), "Images_Print/mr_{}.png".format(index))
print("index", index, "Completed")
return {'ct': ct_img, 'mr': mr_img, 'ct_label': ct_label, 'mr_label': mr_label}
def __len__(self):
"""Return the total number of images in the dataset.
As we have two datasets with potentially different number of images,
we take a maximum of
"""
return max(self.ct_size, self.mr_size)
if __name__ == '__main__':
counter = 0
dataroot = sys.argv[1]
test = sys.argv[2] == "test"
num_iters = int(sys.argv[3])
only_resize = sys.argv[2] == "only_resize"
if test:
print("Test run")
if only_resize:
print("Only resize")
dataset = DataLoaderDataset(dataroot, no_aug=True, test=test, only_resize=only_resize) # create a dataset given opt.dataset_mode and other options
dataset_size = len(dataset) # get the number of images in the dataset.
print('The number of training images = %d' % dataset_size)
foler_name = sys.argv[4]
if test:
os.makedirs(os.path.join(foler_name, 'testct'), exist_ok = True)
os.makedirs(os.path.join(foler_name, 'testct_labels'), exist_ok = True)
os.makedirs(os.path.join(foler_name, 'testmr'), exist_ok = True)
os.makedirs(os.path.join(foler_name, 'testmr_labels'), exist_ok = True)
else:
os.makedirs(os.path.join(foler_name, 'trainct'), exist_ok = True)
os.makedirs(os.path.join(foler_name, 'trainct_labels'), exist_ok = True)
os.makedirs(os.path.join(foler_name, 'trainmr'), exist_ok = True)
os.makedirs(os.path.join(foler_name, 'trainmr_labels'), exist_ok = True)
for epoch in tqdm(range(1, num_iters+1)): # outer loop for different epochs; we save the model by <epoch_count>, <epoch_count>+<save_latrain_freq>
for i, data in enumerate(dataset): # inner loop within one epoch
if test:
np.savez_compressed(os.path.join(foler_name, 'testct', 'ct_test_{}_image.npz'.format(counter)), data["ct"])
np.savez_compressed(os.path.join(foler_name, 'testct_labels', 'ct_test_{}_label.npz'.format(counter)), data["ct_label"])
np.savez_compressed(os.path.join(foler_name, 'testmr', 'mr_test_{}_image.npz'.format(counter)), data["mr"])
np.savez_compressed(os.path.join(foler_name, 'testmr_labels', 'mr_test_{}_label.npz'.format(counter)), data["mr_label"])
else:
np.savez_compressed(os.path.join(foler_name, 'trainct', 'ct_train_{}_image.npz'.format(counter)), data["ct"])
np.savez_compressed(os.path.join(foler_name, 'trainct_labels', 'ct_train_{}_label.npz'.format(counter)), data["ct_label"])
np.savez_compressed(os.path.join(foler_name, 'trainmr', 'mr_train_{}_image.npz'.format(counter)), data["mr"])
np.savez_compressed(os.path.join(foler_name, 'trainmr_labels', 'mr_train_{}_label.npz'.format(counter)), data["mr_label"])
counter += 1