-
Notifications
You must be signed in to change notification settings - Fork 0
/
data_process.py
219 lines (199 loc) · 9.32 KB
/
data_process.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
import os
import argparse
import pickle
import cv2
import numpy as np
import torch.nn as nn
import torch
import torch.nn.functional as F
from PIL import Image
# from ruamel.yaml import safe_load
from torchvision.transforms import Grayscale, Normalize, ToTensor
from utils.helpers import dir_exists, remove_files
def data_process(data_path, name, patch_size, stride, mode):
save_path = os.path.join(data_path, f"{mode}_pro")
dir_exists(save_path)
remove_files(save_path)
if name == "DRIVE":
img_path = os.path.join(data_path, mode, "images")
gt_path = os.path.join(data_path, mode, "1st_manual")
# dgt_path = os.path.join(data_path, mode, "1st_manual_dilation")
file_list = list(sorted(os.listdir(img_path)))
elif name == "CHASEDB1":
file_list = list(sorted(os.listdir(data_path)))
elif name == "STARE":
img_path = os.path.join(data_path, "stare-images")
gt_path = os.path.join(data_path, "labels-ah")
file_list = list(sorted(os.listdir(img_path)))
elif name == "DCA1":
data_path = os.path.join(data_path, "Database_134_Angiograms")
file_list = list(sorted(os.listdir(data_path)))
elif name == "CHUAC":
img_path = os.path.join(data_path, "Original")
gt_path = os.path.join(data_path, "Photoshop")
file_list = list(sorted(os.listdir(img_path)))
img_list = []
gt_list = []
# dgt_list = []
for i, file in enumerate(file_list):
if name == "DRIVE":
img = Image.open(os.path.join(img_path, file))
gt = Image.open(os.path.join(gt_path, file[0:2] + "_manual1.gif"))
# dgt = Image.open(os.path.join(dgt_path, file[0:2] + "_manual1.PNG"))
# img = Grayscale(1)(img)
img_list.append(ToTensor()(img))
gt_list.append(ToTensor()(gt))
# dgt_list.append(ToTensor()(dgt))
elif name == "CHASEDB1":
if len(file) == 13:
if mode == "training" and int(file[6:8]) <= 10:
img = Image.open(os.path.join(data_path, file))
gt = Image.open(os.path.join(
data_path, file[0:9] + '_1stHO.png'))
img = Grayscale(1)(img)
img_list.append(ToTensor()(img))
gt_list.append(ToTensor()(gt))
elif mode == "test" and int(file[6:8]) > 10:
img = Image.open(os.path.join(data_path, file))
gt = Image.open(os.path.join(
data_path, file[0:9] + '_1stHO.png'))
img = Grayscale(1)(img)
img_list.append(ToTensor()(img))
gt_list.append(ToTensor()(gt))
elif name == "DCA1":
if len(file) <= 7:
if mode == "training" and int(file[:-4]) <= 100:
img = cv2.imread(os.path.join(data_path, file), 0)
gt = cv2.imread(os.path.join(
data_path, file[:-4] + '_gt.pgm'), 0)
gt = np.where(gt >= 100, 255, 0).astype(np.uint8)
img_list.append(ToTensor()(img))
gt_list.append(ToTensor()(gt))
elif mode == "test" and int(file[:-4]) > 100:
img = cv2.imread(os.path.join(data_path, file), 0)
gt = cv2.imread(os.path.join(
data_path, file[:-4] + '_gt.pgm'), 0)
gt = np.where(gt >= 100, 255, 0).astype(np.uint8)
img_list.append(ToTensor()(img))
gt_list.append(ToTensor()(gt))
elif name == "CHUAC":
if mode == "training" and int(file[:-4]) <= 20:
img = cv2.imread(os.path.join(img_path, file), 0)
if int(file[:-4]) <= 17 and int(file[:-4]) >= 11:
tail = "PNG"
else:
tail = "png"
gt = cv2.imread(os.path.join(
gt_path, "angio"+file[:-4] + "ok."+tail), 0)
gt = np.where(gt >= 100, 255, 0).astype(np.uint8)
img = cv2.resize(
img, (512, 512), interpolation=cv2.INTER_LINEAR)
cv2.imwrite(f"save_picture/{i}img.png", img)
cv2.imwrite(f"save_picture/{i}gt.png", gt)
img_list.append(ToTensor()(img))
gt_list.append(ToTensor()(gt))
elif mode == "test" and int(file[:-4]) > 20:
img = cv2.imread(os.path.join(img_path, file), 0)
gt = cv2.imread(os.path.join(
gt_path, "angio"+file[:-4] + "ok.png"), 0)
gt = np.where(gt >= 100, 255, 0).astype(np.uint8)
img = cv2.resize(
img, (512, 512), interpolation=cv2.INTER_LINEAR)
cv2.imwrite(f"save_picture/{i}img.png", img)
cv2.imwrite(f"save_picture/{i}gt.png", gt)
img_list.append(ToTensor()(img))
gt_list.append(ToTensor()(gt))
elif name == "STARE":
if not file.endswith("gz"):
img = Image.open(os.path.join(img_path, file))
gt = Image.open(os.path.join(gt_path, file[0:6] + '.ah.ppm'))
cv2.imwrite(f"save_picture/{i}img.png", np.array(img))
cv2.imwrite(f"save_picture/{i}gt.png", np.array(gt))
img = Grayscale(1)(img)
img_list.append(ToTensor()(img))
gt_list.append(ToTensor()(gt))
img_list = normalization(img_list)
if mode == "test":
img_patch = get_patch(img_list, patch_size, stride)
gt_patch = get_patch(gt_list, patch_size, stride)
dgt_patch = get_patch(dgt_list, patch_size, stride)
cgt_patch = get_patch(cgt_list, patch_size, stride)
save_patch(img_patch, save_path, "img_patch", name)
save_patch(gt_patch, save_path, "gt_patch", name)
save_patch(dgt_patch, save_path, "dgt_patch", name)
save_patch(cgt_patch, save_path, "cgt_patch", name)
elif mode == "training":
if name != "CHUAC":
img_list = get_square(img_list, name)
gt_list = get_square(gt_list, name)
dgt_list = get_square(dgt_list, name)
cgt_list = get_square(cgt_list, name)
save_each_image(img_list, save_path, "img", name)
save_each_image(gt_list, save_path, "gt", name)
save_each_image(dgt_list, save_path, "dgt", name)
save_each_image(cgt_list, save_path, "cgt", name)
def get_square(img_list, name):
img_s = []
if name == "DRIVE":
shape = 608
elif name == "CHASEDB1":
shape = 1008
elif name == "DCA1":
shape = 320
_, h, w = img_list[0].shape
pad = nn.ConstantPad2d((0, shape-w, 0, shape-h), 0)
for i in range(len(img_list)):
img = pad(img_list[i])
img_s.append(img)
return img_s
def get_patch(imgs_list, patch_size, stride):
image_list = []
_, h, w = imgs_list[0].shape
pad_h = stride - (h - patch_size) % stride
pad_w = stride - (w - patch_size) % stride
for sub1 in imgs_list:
image = F.pad(sub1, (0, pad_w, 0, pad_h), "constant", 0)
image = image.unfold(1, patch_size, stride).unfold(
2, patch_size, stride).permute(1, 2, 0, 3, 4)
image = image.contiguous().view(
image.shape[0] * image.shape[1], image.shape[2], patch_size, patch_size)
for sub2 in image:
image_list.append(sub2)
return image_list
def save_patch(imgs_list, path, type, name):
for i, sub in enumerate(imgs_list):
with open(file=os.path.join(path, f'{type}_{i}.pkl'), mode='wb') as file:
pickle.dump(np.array(sub), file)
print(f'save {name} {type} : {type}_{i}.pkl')
def save_each_image(imgs_list, path, type, name):
for i, sub in enumerate(imgs_list):
with open(file=os.path.join(path, f'{type}_{i}.pkl'), mode='wb') as file:
pickle.dump(np.array(sub), file)
print(f'save {name} {type} : {type}_{i}.pkl')
def normalization(imgs_list):
imgs = torch.cat(imgs_list, dim=0)
mean = torch.mean(imgs)
std = torch.std(imgs)
normal_list = []
for i in imgs_list:
n = Normalize([mean], [std])(i)
n = (n - torch.min(n)) / (torch.max(n) - torch.min(n))
normal_list.append(n)
return normal_list
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('-dp', '--dataset_path', default="/root/FR-UNet-master/dataset/DRIVE", type=str,
help='the path of dataset',required=False)
parser.add_argument('-dn', '--dataset_name', default="DRIVE", type=str,
help='the name of dataset',choices=['DRIVE','CHASEDB1','STARE','CHUAC','DCA1'],required=False)
parser.add_argument('-ps', '--patch_size', default=48,
help='the size of patch for image partition')
parser.add_argument('-s', '--stride', default=6,
help='the stride of image partition')
args = parser.parse_args()
# with open('config.yaml', encoding='utf-8') as file:
# CFG = safe_load(file) # 为列表类型
data_process(args.dataset_path, args.dataset_name,
args.patch_size, args.stride, "training")
# data_process(args.dataset_path, args.dataset_name,
# args.patch_size, args.stride, "test")