-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathutils.py
240 lines (205 loc) · 8.79 KB
/
utils.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
import torch
import torch.nn as nn
from torch.autograd import Variable
import torch.nn.functional as F
import matplotlib.pyplot as plt
import numpy as np
#%matplotlib inline
import matplotlib.pyplot as plt
import matplotlib
def truncated_normal_(tensor, mean=0, std=1):
size = tensor.shape
tmp = tensor.new_empty(size + (4,)).normal_()
valid = (tmp < 2) & (tmp > -2)
ind = valid.max(-1, keepdim=True)[1]
tensor.data.copy_(tmp.gather(-1, ind).squeeze(-1))
tensor.data.mul_(std).add_(mean)
def init_weights(m):
if type(m) == nn.Conv2d or type(m) == nn.ConvTranspose2d:
nn.init.kaiming_normal_(m.weight, mode='fan_in', nonlinearity='relu')
#nn.init.normal_(m.weight, std=0.001)
#nn.init.normal_(m.bias, std=0.001)
truncated_normal_(m.bias, mean=0, std=0.001)
def init_weights_orthogonal_normal(m):
if type(m) == nn.Conv2d or type(m) == nn.ConvTranspose2d:
nn.init.orthogonal_(m.weight)
truncated_normal_(m.bias, mean=0, std=0.001)
#nn.init.normal_(m.bias, std=0.001)
def l2_regularisation(m):
l2_reg = None
for W in m.parameters():
if l2_reg is None:
l2_reg = W.norm(2)
else:
l2_reg = l2_reg + W.norm(2)
return l2_reg
def save_mask_prediction_example(mask, pred, iter):
plt.imshow(pred[0,:,:],cmap='Greys')
plt.savefig('images/'+str(iter)+"_prediction.png")
plt.imshow(mask[0,:,:],cmap='Greys')
plt.savefig('images/'+str(iter)+"_mask.png")
def label2multichannel(mask, class_num=10):
"""将单通道（元素值1-10）变为多通道（元素值0-1）
输入为torch
输出为torch
mask.shape: batch_size,1,240,240
"""
mask = mask.numpy()
batch_size = mask.shape[0]
h, w = mask.shape[2], mask.shape[3]
# mask = mask.reshape((240, 240))
label = np.zeros((batch_size,class_num, 240, 240))
for b in range(batch_size):
for i in range(class_num):
for x in range(h):
for y in range(w):
if int(mask[b,0,x,y]) == i+1:
label[b,i,x,y] = 1
label = torch.from_numpy(label)
label = label.type(torch.FloatTensor)
return label
def mask2rgb(mask, class_num=10):
"""单通道变rgb图，以不同颜色显示大脑不同区域
输入numpy(1,240,240)/(240,240)
输出numpy(3,240,240)
mask: numpy
mask.shape (240,240)
"""
mask = mask.reshape((240,240))
# 颜色rgb （黑，红，绿，蓝，黄，淡紫，青，紫，黄绿，白）
color = [(0,0,0),(255,0,0),(0,255,0),(0,0,255),(255,255,0),
(255,0,255),(0,255,255),(128,0,128),(128,128,0),(255,255,255)]
img = np.zeros((240, 240, 3))
for c in range(class_num):
for x in range(240):
for y in range(240):
if int(mask[x,y]) == c:
img[x,y,0] = color[c][0]
img[x,y,1] = color[c][1]
img[x,y,2] = color[c][2]
img /= 255
return img
def cal_variance(image_np, label_np, mask_pros, mask_pres, class_num, series_uid):
"""计算多个预测结果的方差
image_np:(240,240)
label_np:(240,240),元素值0-8
mask_pres:(m,class_num,240,240),元素值为预测结果, np
mask_pros:(m,240,240),m为预测次数,元素值0-8, list(np)
class_num: 分割任务类别数
series_uid: 每张图片的序列号，list，[病人，张数]
"""
m = len(mask_pros) # 预测结果（次）数
h, w = image_np.shape # 图片高宽
## 方法二（对预测正确的，对0/1求方差）
# mean_result = np.zeros((h, w)) # 均值
# variance_result = np.zeros((h, w)) # 方差
# mask_pros_temp = [np.zeros((240, 240)) for i in range(m)] # 记录预测正确的像素点设为1
# for i in range(m):
# mask_pros_temp[i] = (mask_pros[i]==label_np)
# # 计算均值
# for i in range(m):
# mean_result += mask_pros_temp[i]
# mean_result /= m
# # 计算方差
# for i in range(m):
# variance_result += np.square(mean_result - mask_pros_temp[i])
# variance_result /= m
## 方法三
mask_pres = np.array(mask_pres) # (m,class_num,240,240)
variance_result = np.var(mask_pres, axis=0)
variance_result = np.sum(variance_result, axis=0)
# 熵
eps = 0.0001
sample = np.mean(mask_pres,axis=0) # (class_num,240,240)
sample += eps
entropy = -np.sum(sample * np.log(sample), axis=0) # (240,240)
# 保存原图、标签、和m张预测结果
# save_8_pred_img(image_np, label_np, variance_result,
# mask_pros, entropy, class_num, series_uid)
# 保存原图、标签、和方差
# save_variance_img(image_np, mask2rgb(label_np), variance_result,series_uid)
return entropy, variance_result
def save_variance_img(orig, mask, var, series_uid):
"""保存原图、标签、预测结果的方差进行对比"""
fig, ax = plt.subplots(1, 3, sharey=True, figsize=(16, 5))
ax[0].set_title("Original")
ax[1].set_title("Ground Truth")
ax[2].set_title("variance")
ax[0].imshow(orig, aspect="auto", cmap="gray")
ax[1].imshow(mask, aspect="auto")
ax[2].imshow(var, aspect="auto")
# 去掉刻度
for i in range(3):
ax[i].axis('off')
fig.suptitle('patient {} - slice {}'.format(series_uid[0][0], series_uid[1][0]))
plt.savefig('picture/p{}_s{}_var.jpg'.format(series_uid[0][0], series_uid[1][0]))
plt.close()
def save_8_pred_img(orig, mask, var, pred, entropy, class_num, series_uid):
"""保存原图、标签、8个预测结果进行对比
orig:(240,240)
mask:(240,240),元素值0-9
pred:(k,240,240),k为预测次数,元素值0-9
"""
m = len(pred)
fig, ax = plt.subplots(5, 4, sharey=True, figsize=(24, 25))
cmap = plt.cm.get_cmap('tab10', 10) # 10 discrete colors,tab10,Paired
# 设置每张子图的标题
ax[0][0].set_title("Original")
ax[0][1].set_title("Ground Truth")
ax[0][2].set_title("Entropy")
ax[0][3].set_title("variance")
for i in range(4):
ax[1][i].set_title("predict_{}".format(i))
ax[2][i].set_title("predict_{}".format(i+4))
ax[3][i].set_title("predict_{}".format(i+8))
ax[4][i].set_title("predict_{}".format(i+12))
# 设置每张子图显示内容
ax00 = ax[0][0].imshow(orig, aspect="auto", cmap="gray")
ax01 = ax[0][1].imshow(mask, cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax02 = ax[0][2].imshow(entropy, aspect="auto", cmap="jet", vmin=0, vmax=2)
ax03 = ax[0][3].imshow(var, aspect="auto", cmap="jet")
# for i in range(4):
ax10 = ax[1][0].imshow(pred[0], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax11 = ax[1][1].imshow(pred[1], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax12 = ax[1][2].imshow(pred[2], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax13 = ax[1][3].imshow(pred[3], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax20 = ax[2][0].imshow(pred[4], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax21 = ax[2][1].imshow(pred[5], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax22 = ax[2][2].imshow(pred[6], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax23 = ax[2][3].imshow(pred[7], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax30 = ax[3][0].imshow(pred[8], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax31 = ax[3][1].imshow(pred[9], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax32 = ax[3][2].imshow(pred[10], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax33 = ax[3][3].imshow(pred[11], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax40 = ax[4][0].imshow(pred[12], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax41 = ax[4][1].imshow(pred[13], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax42 = ax[4][2].imshow(pred[14], cmap=cmap, aspect="auto", vmin=0, vmax=9)
ax43 = ax[4][3].imshow(pred[15], cmap=cmap, aspect="auto", vmin=0, vmax=9)
# color bar
fig.colorbar(ax00, ax=ax[0][0])
fig.colorbar(ax01, ax=ax[0][1])
fig.colorbar(ax02, ax=ax[0][2])
fig.colorbar(ax03, ax=ax[0][3])
fig.colorbar(ax10, ax=ax[1][0])
fig.colorbar(ax11, ax=ax[1][1])
fig.colorbar(ax12, ax=ax[1][2])
fig.colorbar(ax13, ax=ax[1][3])
fig.colorbar(ax20, ax=ax[2][0])
fig.colorbar(ax21, ax=ax[2][1])
fig.colorbar(ax22, ax=ax[2][2])
fig.colorbar(ax23, ax=ax[2][3])
fig.colorbar(ax30, ax=ax[3][0])
fig.colorbar(ax31, ax=ax[3][1])
fig.colorbar(ax32, ax=ax[3][2])
fig.colorbar(ax33, ax=ax[3][3])
fig.colorbar(ax40, ax=ax[4][0])
fig.colorbar(ax41, ax=ax[4][1])
fig.colorbar(ax42, ax=ax[4][2])
fig.colorbar(ax43, ax=ax[4][3])
# 不显示刻度尺、标
for i in range(5):
for j in range(4):
ax[i][j].axis('off')
fig.suptitle('patient {} - slice {}'.format(series_uid[0][0], series_uid[1][0]))
plt.savefig('picture/c{}_p{}_s{}_pre{}.jpg'.format(class_num, series_uid[0][0], series_uid[1][0], m))
plt.close()