-
Notifications
You must be signed in to change notification settings - Fork 89
/
simple_mixup.py
48 lines (38 loc) · 1.4 KB
/
simple_mixup.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
import numpy as np
import cv2
"""
说明:
这里是前期看到公式后一个粗略的实现
参考了更多资料后其实有些地方还是有点问题
比如weight应该是一个beta分布而不是均匀分布,如
weight = np.random.beta(alpha,alpha)
然后是输入每个batch的x,y即可,统一和一个从数据集中随机选取的图片做mixup
最后,最关键的是,最后计算loss并不是修改标签y,
而是如下计算loss:
loss = weight * criterion(outputs, targets_a) + (1 - weight) * criterion(outputs, targets_b);
下一次用到mixup的时候修改了再更新到这里吧。
5.24: 已更新beta分布
"""
def mixup_batch(x1,y1,x2,y2,alpha=0.4):
"""
get batch data
:param x: two training imgs (same shape) ndarry
:param y: two one-hot labels(same shape) ndarry
:param alpha: hyper-parameter α, default as 0.2
:return: new_x,new_y
"""
#weight = np.random.choice([0.1,0.2,0.3,0.4,0.6,0.7,0.8,0.9])
weight = np.random.beta(alpha, alpha)
print(weight)
new_x = x1*weight+x2*(1-weight)
new_y = y1*weight+y2*(1-weight)
return new_x, new_y
img1 = cv2.imread("1.jpg")
img1 = cv2.resize(img1,(224,224))
img2 = cv2.imread("2.jpg")
img2 = cv2.resize(img2,(224,224))
y1 = np.array([0,0,1])
y2 = np.array([0,1,0])
x,y = mixup_batch(img1,y1,img2,y2)
cv2.imwrite("12.jpg", x)
print(y)