-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathgencutimg_segformer.py
194 lines (161 loc) · 7.64 KB
/
gencutimg_segformer.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
import numpy as np
import pandas as pd
import os
import rasterio
from torch.utils.data import Dataset
from rasterio.windows import Window
import cv2
from tqdm import tqdm
from PIL import Image
import tifffile
from tqdm.contrib import tzip
from scipy.ndimage.morphology import distance_transform_edt
def read_tiff(path, scale=None, verbose=0):
image = tifffile.imread(path)
if len(image.shape) == 5:
image = image.squeeze().transpose(1, 2, 0)
if verbose:
print(f"[{path}] Image shape: {image.shape}")
if scale:
new_size = (image.shape[1] // scale, image.shape[0] // scale)
image = cv2.resize(image, new_size)
if verbose:
print(f"[{path}] Resized Image shape: {image.shape}")
return image
def onehot_to_multiclass_edges(mask, radius, num_classes):
"""
Converts a segmentation mask (K,H,W) to an edgemap (K,H,W)
"""
if radius < 0:
return mask
# We need to pad the borders for boundary conditions
mask_pad = np.pad(mask, ((0, 0), (1, 1), (1, 1)), mode='constant', constant_values=0)
channels = []
for i in range(num_classes):
dist = distance_transform_edt(mask_pad[i, :]) + distance_transform_edt(1.0 - mask_pad[i, :])
dist = dist[1:-1, 1:-1]
dist[dist > radius] = 0
dist = (dist > 0).astype(np.uint8)
channels.append(dist)
return np.array(channels)
def onehot_to_binary_edges(mask, radius, num_classes):
"""
Converts a segmentation mask (K,H,W) to a binary edgemap (H,W)
"""
if radius < 0:
return mask
# We need to pad the borders for boundary conditions
mask_pad = np.pad(mask, ((0, 0), (1, 1), (1, 1)), mode='constant', constant_values=0)
edgemap = np.zeros(mask.shape[1:])
for i in range(num_classes):
# ti qu lun kuo
dist = distance_transform_edt(mask_pad[i, :]) + distance_transform_edt(1.0 - mask_pad[i, :])
dist = dist[1:-1, 1:-1]
dist[dist > radius] = 0
edgemap += dist
# edgemap = np.expand_dims(edgemap, axis=0)
edgemap = (edgemap > 0).astype(np.uint8)*255
return edgemap
def mask_to_onehot(mask, num_classes):
"""
Converts a segmentation mask (H,W) to (K,H,W) where the last dim is a one
hot encoding vector
"""
_mask = [mask == (i) for i in range(num_classes)]
return np.array(_mask).astype(np.uint8)
class HuBMAPDataset(Dataset):
def __init__(self, img_idx, mask_idx, organ):
self.img = cv2.cvtColor(read_tiff(img_idx), cv2.COLOR_RGB2BGR)
self.mask = cv2.imread(mask_idx, cv2.IMREAD_GRAYSCALE)
self.organ = organ
def __len__(self):
return 1
def __getitem__(self, idx):
# if self.organ == 'prostate':
# self.img = cv2.resize(self.img, (190, 190),
# interpolation=cv2.INTER_AREA)
# self.img = cv2.resize(self.img, (190, 190),
# interpolation=cv2.INTER_AREA)
# self.mask = cv2.resize(self.mask, (190, 190),
# interpolation=cv2.INTER_NEAREST)
img = cv2.resize(self.img, (img_size, img_size),
interpolation=cv2.INTER_AREA)
mask = cv2.resize(self.mask, (mask_size, mask_size),
interpolation=cv2.INTER_NEAREST)
return img, mask, 1
for K in [1, 2, 3, 4, 5]:
for mode in ['train', 'valid']:
MASKS = r'/root/autodl-tmp/train.csv'
DATA = r'/root/autodl-tmp/train_images'
MASK = r'/root/autodl-tmp/mask'
path = r'/root/autodl-tmp'
DATA_cut = f'/root/autodl-tmp/{mode}_data_cut_768_x3_fold_{K}_lung'
MASK_cut = f'/root/autodl-tmp/{mode}_mask_cut_768_x3_fold_{K}_lung'
EDGE_cut = f'/root/autodl-tmp/{mode}_edge_cut_768_x3_fold_{K}_lung'
img_txt = f'./train_valid_list_path/{mode}_img_fold_{K}.txt'
mask_txt = f'./train_valid_list_path/{mode}_mask_fold_{K}.txt'
data_cut_txt = f"{mode}_img_768_x3_fold_{K}_lung.txt"
mask_cut_txt = f"{mode}_mask_768_x3_fold_{K}_lung.txt"
edge_cut_txt = f"{mode}_edge_768_x3_fold_{K}_lung.txt"
if mode == 'train':
img_size = 768
mask_size = 768
if mode == 'valid':
img_size = 768
mask_size = 768
if not os.path.exists(EDGE_cut):
os.mkdir(EDGE_cut)
if not os.path.exists(DATA_cut):
os.mkdir(DATA_cut)
os.mkdir(MASK_cut)
f = open(img_txt, 'r')
img_list = list(f)
f.close()
f = open(mask_txt, 'r')
mask_list = list(f)
f.close()
df_masks = pd.read_csv(MASKS)[['id', 'rle']].set_index('id')
df_masks.head()
df = pd.read_csv(os.path.join(MASKS))
if mode == 'valid':
with open(os.path.join(path, data_cut_txt), 'w') as train, open(os.path.join(path, mask_cut_txt), 'w') as test:
for img_index, mask_index in tzip(img_list, mask_list):
img_index = img_index.strip('\n')
mask_index = mask_index.strip('\n')
index = int(img_index.split("images/")[1].split(".tiff")[0])
organ = df[df["id"] == index]["organ"].iloc[-1]
if organ != 'lung':
continue
ds = HuBMAPDataset(img_idx=img_index, mask_idx=mask_index, organ=organ)
for i in range(len(ds)):
im, m, idx = ds[i]
if idx < 0: continue
cv2.imwrite(os.path.join(DATA_cut, f'{index}_{idx:04d}.png'), im)
cv2.imwrite(os.path.join(MASK_cut, f'{index}_{idx:04d}.png'), m)
train.write(os.path.join(DATA_cut, f'{index}_{idx:04d}.png') + '\n')
test.write(os.path.join(MASK_cut, f'{index}_{idx:04d}.png') + '\n')
if mode == 'train':
with open(os.path.join(path, data_cut_txt), 'w') as train, open(os.path.join(path, mask_cut_txt), 'w') as test, open(os.path.join(path, edge_cut_txt), 'w') as edge:
for img_index, mask_index in tzip(img_list, mask_list):
img_index = img_index.strip('\n')
mask_index = mask_index.strip('\n')
index = int(img_index.split("images/")[1].split(".tiff")[0])
organ = df[df["id"] == index]["organ"].iloc[-1]
if organ != 'lung':
continue
ds = HuBMAPDataset(img_idx=img_index, mask_idx=mask_index, organ=organ)
for i in range(len(ds)):
im, m, idx = ds[i]
if idx < 0: continue
oneHot_label = mask_to_onehot(m, 2)
edgee = onehot_to_binary_edges(oneHot_label, 2, 2)
edgee[:2, :] = 0
edgee[-2:, :] = 0
edgee[:, :2] = 0
edgee[:, -2:] = 0
cv2.imwrite(os.path.join(DATA_cut, f'{index}_{idx:04d}.png'), im)
cv2.imwrite(os.path.join(MASK_cut, f'{index}_{idx:04d}.png'), m)
cv2.imwrite(os.path.join(EDGE_cut, f'{index}_{idx:04d}.png'), edgee)
train.write(os.path.join(DATA_cut, f'{index}_{idx:04d}.png') + '\n')
test.write(os.path.join(MASK_cut, f'{index}_{idx:04d}.png') + '\n')
edge.write(os.path.join(EDGE_cut, f'{index}_{idx:04d}.png') + '\n')