-
Notifications
You must be signed in to change notification settings - Fork 2
/
load_polyp_data.py
270 lines (237 loc) · 12.5 KB
/
load_polyp_data.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
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
'''
Diagnosing Colorectal Polyps in the Wild with Capsule Networks (D-Caps)
Original Paper by Rodney LaLonde, Pujan Kandel, Concetto Spampinato, Michael B. Wallace, and Ulas Bagci
Paper published at ISBI 2020: arXiv version (https://arxiv.org/abs/2001.03305)
Code written by: Rodney LaLonde
If you use significant portions of this code or the ideas from our paper, please cite it :)
If you have any questions, please email me at [email protected].
This file handles everything data related: Loading the data, splitting it, etc.
'''
from __future__ import print_function
from collections import Counter
import os
from glob import glob
import csv
import cv2
from sklearn.model_selection import StratifiedKFold
import numpy as np
from sklearn.model_selection import train_test_split
from skimage.transform import resize
from tqdm import tqdm
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
plt.ioff()
from utils import safe_mkdir
debug = False
def load_data(root, exp_name, exp=0, split=0, k_folds=4, val_split=0.1):
# Main functionality of loading and spliting the data
def _load_data():
with open(os.path.join(root, 'split_lists', exp_name, 'train_split_' + str(split) + '.csv'), 'r') as f:
reader = csv.reader(f)
training_list = list(reader)
with open(os.path.join(root, 'split_lists', exp_name, 'test_split_' + str(split) + '.csv'), 'r') as f:
reader = csv.reader(f)
test_list = list(reader)
X, y = np.hsplit(np.asarray(training_list), 2)
X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=val_split, random_state=12, stratify=y)
new_train_list, val_list = np.hstack((X_train, y_train)), np.hstack((X_val, y_val))
return new_train_list, val_list, np.asarray(test_list)
# Try-catch to handle calling split data before load only if files are not found.
try:
new_training_list, validation_list, testing_list = _load_data()
return new_training_list, validation_list, testing_list
except:
# Create the training and test splits if not found
split_data(root, exp_name, exp_num=exp, num_splits=k_folds)
try:
new_training_list, validation_list, testing_list = _load_data()
return new_training_list, validation_list, testing_list
except Exception as e:
print(e)
print('Failed to load data, see load_data in load_polyp_data.py')
exit(1)
def split_data_for_flow(root, out_dir, exp_name, resize_option, resize_shape, train_list, val_list, test_list):
def _load_imgs(data_list, phase):
data_info_list = []
for patient_num_label in tqdm(data_list, desc=phase):
files = []
for ext in ('*.jpg', '*.JPG', '*.tif', '*.tiff', '*.png', '*.PNG'):
files.extend(sorted(glob(os.path.join(root, patient_num_label[0], ext).replace('\\', '/'))))
if not files:
print('WARNING: No Images found in {}. Ensure the path is set up properly in the compute '
'class samples function in load polyp data.'.format(os.path.join(root, patient_num_label[0])))
for f in files:
img = cv2.imread(f)
try:
img = img.astype(np.float32)
except:
print('Unable to load image: {}. Please check the file for corruption.'.format(f))
continue
data_info_list.append([f, patient_num_label[1], img.shape[0], img.shape[1]])
# Balance sample amounts
np_data_list = np.asarray(data_info_list)
if phase == 'Load_train':
n_classes = len(np.unique(np_data_list[:,1]))
max_samples = 0
split_np_list = []
for n in range(n_classes):
split_np_list.append(np_data_list[np_data_list[:, 1] == '{}'.format(n)])
amt = len(split_np_list[n])
if amt > max_samples:
max_samples = amt
out_list = np.empty((n_classes * max_samples,5), dtype='|S255')
for n in range(n_classes):
res_lis = np.resize(split_np_list[n],(max_samples,4))
out_list[n*max_samples:(n+1)*max_samples,:] = np.hstack((res_lis, np.expand_dims(res_lis[:,0],-1)))
counts = Counter(out_list[:, 4])
for s, num in tqdm(counts.items(), desc='Renaming duplicate images'):
if num > 1: # ignore strings that only appear once
for suffix in range(1, num + 1): # suffix starts at 1 and increases by 1 each time
out_list[out_list[:, 4].tolist().index(s), 4] = \
'{}_{}.{}'.format(s.decode('utf-8').replace('\\', '/')[:-4],
suffix, s.decode('utf-8').replace('\\', '/')[-3:]) # replace each appearance of s
return out_list
else:
return np.hstack((np_data_list, np.expand_dims(np_data_list[:,0],-1)))
def _compute_out_shape(height_list, width_list):
if resize_option == 'resize_max':
out_shape = [np.max(height_list), np.max(width_list)]
elif resize_option == 'resize_min' or resize_option == 'crop_min':
out_shape = [np.min(height_list), np.min(width_list)]
elif resize_option == 'resize_avg':
out_shape = [int(np.mean(height_list)), int(np.mean(width_list))]
elif resize_option == 'resize_std':
out_shape = resize_shape
else:
raise NotImplementedError(
'Error: Encountered resize choice which is not implemented in load_polyp_data.py.')
if resize_shape[0] is not None:
out_shape[0] = resize_shape[0]
if resize_shape[1] is not None:
out_shape[1] = resize_shape[1]
return out_shape[0], out_shape[1]
def _random_crop(img, crop_shape, mask=None):
assert img.shape[2] == 3
height, width = img.shape[0], img.shape[1]
dy, dx = crop_shape
if dy is None:
dy = height
if dx is None:
dx = width
x = np.random.randint(0, width - dx + 1)
y = np.random.randint(0, height - dy + 1)
if mask is not None:
return [img[y:(y+dy), x:(x+dx), :], mask[y:(y+dy), x:(x+dx)]]
else:
return [img[y:(y + dy), x:(x + dx), :]]
def _save_imgs(lst, phase, hei, wid):
class_list = exp_name.split('vs')
class_map = dict()
for k, v in enumerate(class_list):
class_map[k] = '{}_{}'.format(k,v)
try:
safe_mkdir(os.path.join(out_dir, phase, class_map[k]))
except:
pass
for i, f in enumerate(tqdm(lst, desc='Creating {} images'.format(phase))):
img_out_name = os.path.join(out_dir, phase, class_map[int(f[1])],
'{}_{}.jpg'.format(os.path.basename(os.path.dirname(f[4])),
os.path.basename(f[4])[:-4])).replace('\\', '/')
if not os.path.isfile(img_out_name):
try:
im = cv2.imread(f[0].decode('utf-8'))
except AttributeError:
im = cv2.imread(f[0])
try:
im = im.astype(np.float32)
except:
print('Unable to load image: {}. Please check the file for corruption.'.format(f[0]))
continue
if im.shape[0] != hei or im.shape[1] != wid:
if resize_option == 'crop_min':
out_im = _random_crop(im, (hei,wid))
else:
out_im = resize(im, (hei,wid), mode='reflect', preserve_range=True)
else:
out_im = im
cv2.imwrite(img_out_name, out_im)
def _compute_num_images():
n_train = len(glob(os.path.join(out_dir, 'train', '*', '*.jpg')))
n_val = len(glob(os.path.join(out_dir, 'val', '*', '*.jpg')))
n_test = len(glob(os.path.join(out_dir, 'test', '*', '*.jpg')))
return n_train, n_val, n_test
train_info_array = np.asarray(_load_imgs(train_list, 'Load_train'))
val_info_array = np.asarray(_load_imgs(val_list, 'Load_val'))
test_info_array = np.asarray(_load_imgs(test_list, 'Load_test'))
train_height, train_width = _compute_out_shape(train_info_array[:,2].astype(int), train_info_array[:,3].astype(int))
val_height, val_width = _compute_out_shape(val_info_array[:,2].astype(int), val_info_array[:,3].astype(int))
test_height, test_width = _compute_out_shape(test_info_array[:,2].astype(int), test_info_array[:,3].astype(int))
_save_imgs(train_info_array, 'train', train_height, train_width)
_save_imgs(val_info_array, 'val', val_height, val_width)
_save_imgs(test_info_array, 'test', test_height, test_width)
num_train, num_val, num_test = _compute_num_images()
return num_train, [train_height, train_width], num_val, [val_height, val_width], \
num_test, [test_height, test_width]
def split_data(root_path, exp_name, exp_num, num_splits=4):
patient_list = []
patient_list.extend(sorted(glob(os.path.join(root_path,'Images','*', '*'))))
assert len(patient_list) != 0, 'Unable to find any files in {}'.format(os.path.join(root_path,'Images','*','*'))
label_list = []
for patient_num in patient_list:
img_type = os.path.basename(os.path.dirname(patient_num))
if img_type == 'Normal':
label_list.append(0)
elif img_type == 'HP' or img_type == 'Hyperplastic':
label_list.append(1)
elif img_type == 'Serrated' or img_type == 'SSA':
label_list.append(2)
elif img_type == 'TA' or img_type == 'TVA' or img_type == 'Adenoma':
label_list.append(3)
elif img_type == 'Cancer':
label_list.append(4)
elif img_type == 'NewAdenomas':
label_list.append(5)
pass # This is a holdout testing set. Do not add to training, val, or testing. Only task after cross-validation is complete.
else:
raise Exception('Encountered unknown image type: {}'.format(img_type))
outdir = os.path.join(root_path, 'split_lists', exp_name)
try:
safe_mkdir(outdir)
except:
pass
patient_list = np.asarray(patient_list)
label_list = np.asarray(label_list)
if exp_num == 0:
to_delete = np.append(np.argwhere(label_list==0), np.append(np.argwhere(label_list==2), np.append(np.argwhere(label_list==4), np.argwhere(label_list==5))))
final_img_list = np.delete(patient_list, to_delete)
final_label_list = np.delete(label_list, to_delete)
final_label_list[final_label_list==1] = 0
final_label_list[final_label_list==3] = 1
elif exp_num == 1:
to_delete = np.append(np.argwhere(label_list==0), np.append(np.argwhere(label_list==4), np.argwhere(label_list==5)))
final_img_list = np.delete(patient_list, to_delete)
final_label_list = np.delete(label_list, to_delete)
final_label_list[final_label_list==1] = 0
final_label_list[final_label_list==2] = 1
final_label_list[final_label_list==3] = 1
elif exp_num == 2:
to_delete = np.append(np.argwhere(label_list==0), np.append(np.argwhere(label_list==3), np.append(np.argwhere(label_list==4), np.argwhere(label_list==5))))
final_img_list = np.delete(patient_list, to_delete)
final_label_list = np.delete(label_list, to_delete)
final_label_list[final_label_list==1] = 0
final_label_list[final_label_list==2] = 1
else:
raise Exception('Experiment number undefined.')
skf = StratifiedKFold(n_splits=num_splits, shuffle=True, random_state=12)
n = 0
for train_index, test_index in skf.split(final_img_list, final_label_list):
with open(os.path.join(outdir,'train_split_' + str(n) + '.csv'), 'w', newline='') as csvfile:
writer = csv.writer(csvfile, delimiter=',', quotechar='|', quoting=csv.QUOTE_MINIMAL)
for i in train_index:
writer.writerow([final_img_list[i].split(root_path)[1].replace('\\', '/'), final_label_list[i]])
with open(os.path.join(outdir,'test_split_' + str(n) + '.csv'), 'w', newline='') as csvfile:
writer = csv.writer(csvfile, delimiter=',', quotechar='|', quoting=csv.QUOTE_MINIMAL)
for i in test_index:
writer.writerow([final_img_list[i].split(root_path)[1].replace('\\', '/'), final_label_list[i]])
n += 1