-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathutils.py
123 lines (104 loc) · 4.78 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
import numpy as np
import json
import math
from imageio import imread
import random
class DataLoader:
def __init__(self,batchSize,inputShape,dataList,guaMaxValue):
self.inputShape=inputShape
self.batchSize=batchSize
self.dataList=dataList
self.guaMaxValue=guaMaxValue
def generator(self):
while(1):
batch=np.random.choice(self.dataList,size=self.batchSize,replace=False)
images=[]
labels=[]
for b in batch:
img=imread(b)
# Ajout de perturbation sur les canaux de couleurs
img = change_channel_ratio(img)
images.append(img)
temp=np.load(b.replace(".jpg",".npy")).astype(int)
np.place(temp,temp==255,self.guaMaxValue)
labels.append(temp)
images=np.array(images)
yield (np.array(images)/255.).astype(np.float32),np.array(labels)
# Data augmentation
def change_channel_ratio(img):
img = img.copy()
# Color channels
ratio = 1+(random.random()-0.5)/10
img[:, :, 0] = np.clip(img[:, :, 0]*ratio, 0, 255).astype(np.uint8)
ratio = 1+(random.random()-0.5)/10
img[:, :, 1] = np.clip(img[:, :, 1]*ratio, 0, 255).astype(np.uint8)
ratio = 1+(random.random()-0.5)/10
img[:, :, 2] = np.clip(img[:, :, 2]*ratio, 0, 255).astype(np.uint8)
# Contrast
factor = 1+((random.random()-0.3)/5)
img = np.clip(128 + factor * img - factor * 128, 0, 255).astype(np.uint8)
return img
def dataAugmentation(images,labels):
newImages=[]
newLabels=[]
for i,img in enumerate(images):
newImages.append(img)
newLabels.append(labels[i])
newImages.append(np.flip(img,axis=0))
newLabels.append(np.flip(labels[i],axis=0))
newImages.append(np.flip(img,axis=1))
newLabels.append(np.flip(labels[i],axis=1))
newImages.append(np.rot90(img,k=1))
newLabels.append(np.rot90(labels[i],k=1))
newImages.append(np.rot90(img,k=2))
newLabels.append(np.rot90(labels[i],k=2))
newImages.append(np.rot90(img,k=3))
newLabels.append(np.rot90(labels[i],k=3))
return np.array(newImages),np.array(newLabels)
def createGaussianLabel(labelPath,inputShape,imageShape,GaussianSize):
x, y = np.meshgrid(np.linspace(-1,1,GaussianSize), np.linspace(-1,1,GaussianSize))
d = np.sqrt(x*x+y*y)
sigma, mu = 0.5, 0.0
gua = np.exp(-( (d-mu)**2 / ( 2.0 * sigma**2 ) ) )*255
with open(labelPath) as f:
label = json.load(f)
img=np.zeros(shape=inputShape,dtype=np.int)
guLabel=np.zeros(shape=inputShape,dtype=np.int)
for d in label:
x=min(max(int(int(d['y'])*(inputShape[0]/imageShape[0])),0),inputShape[0])
y=min(max(int(int(d['x'])*(inputShape[1]/imageShape[1])),0),inputShape[1])
x_,y_=img[max(int(x-math.floor(GaussianSize/2)),0):min(int(x+math.ceil(GaussianSize/2)),255),
max(int(y-math.floor(GaussianSize/2)),0):min(int(y+math.ceil(GaussianSize/2)),255),0].shape
if(int(d['label_id'])==1):
guLabel[max(int(x-math.floor(GaussianSize/2)),0):min(int(x+math.ceil(GaussianSize/2)),255),
max(0,int(y-math.floor(GaussianSize/2))):min(int(y+math.ceil(GaussianSize/2)),255),0]=gua[
math.floor(GaussianSize/2)-x_//2:math.floor(GaussianSize/2)+x_//2+x_%2,
math.floor(GaussianSize/2)-y_//2:math.floor(GaussianSize/2)+y_//2+y_%2]
if(int(d['label_id'])==2):
guLabel[max(int(x-math.floor(GaussianSize/2)),0):min(int(x+math.ceil(GaussianSize/2)),255),
max(0,int(y-math.floor(GaussianSize/2))):min(int(y+math.ceil(GaussianSize/2)),255),1]=gua[
math.floor(GaussianSize/2)-x_//2:math.floor(GaussianSize/2)+x_//2+x_%2,
math.floor(GaussianSize/2)-y_//2:math.floor(GaussianSize/2)+y_//2+y_%2]
if(int(d['label_id'])==3):
print('Immune cells')
guLabel[max(int(x-math.floor(GaussianSize/2)),0):min(int(x+math.ceil(GaussianSize/2)),255),
max(0,int(y-math.floor(GaussianSize/2))):min(int(y+math.ceil(GaussianSize/2)),255),1]=gua[
math.floor(GaussianSize/2)-x_//2:math.floor(GaussianSize/2)+x_//2+x_%2,
math.floor(GaussianSize/2)-y_//2:math.floor(GaussianSize/2)+y_//2+y_%2]
return guLabel
class LrPolicy:
def __init__(self,lr):
self.lr=lr
def stepDecay(self,epoch):
step = 10
num = epoch // step
if epoch>=30:
lrate = self.lr/1000
elif num % 3 == 0:
lrate = self.lr
elif num % 3 == 1:
lrate = self.lr/100
else :
lrate = self.lr/1000
print('Learning rate for epoch {} is {}.'.format(epoch+1, lrate))
return np.float(lrate)