-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdata_augmentation.py
156 lines (143 loc) · 5.38 KB
/
data_augmentation.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
import torch
import numpy as np
from PIL import Image
import random
import cv2
def crop(image, heatmap):
image = np.array(image)
heatmap = np.array(heatmap)
heatmap = cv2.cvtColor(heatmap, cv2.COLOR_BGR2GRAY)
gt = heatmap
height, width, _ = image.shape
SCALE = [1.,0.9, 0.85, 0.8,0.75,0.7, 0.6, 0.65, 0.5, 0.55, 0.4, 0.45, 0.3, 0.35, 0.2, 0.1]
# print(heatmap.max(), '1st max')
# print(heatmap.min(), '1st min')
for _ in range(1000):
scale = random.randrange(len(SCALE))
scale = SCALE[scale]
short_side = min(height, width)
w = min(int(scale*short_side) - int(scale*short_side)%4,width)
h = w
l = random.randrange(0,width-w + 1)
left = l - (l%4)
t = random.randrange(0,height-h + 1) # changed height-t to height-h
top = t - (t%4)
crop_rgn = [left, top, left+w, top+h]
crop_im = image[crop_rgn[1]:crop_rgn[3], crop_rgn[0]:crop_rgn[2]]
heatmap = heatmap[int(crop_rgn[1]):int(crop_rgn[3]), int(crop_rgn[0]):int(crop_rgn[2])]
c0, c1 = np.where(heatmap>=250)
# print(heatmap.max(), '2nd max')
# print(heatmap.min(), '2nd min')
# print(c0, 'c0')
# print(c1, 'c1')
# exit(1)
if len(c0) >= 1:
#crop_im = cv2.resize(crop_im,(800,800),cv2.INTER_LINEAR)
#hmap_img = cv2.resize(heatmap.astype(np.float32),(800,800),cv2.INTER_NEAREST)
hmap_img = cv2.cvtColor(heatmap, cv2.COLOR_GRAY2BGR)
return crop_im, hmap_img
else:
heatmap = gt
#heatmap[np.where(heatmap>0)]=1
continue
#image = cv2.resize(image, (800, 800), cv2.INTER_LINEAR)
#heatmap = cv2.resize(heatmap.astype(np.float32), (800, 800), cv2.INTER_NEAREST)
heatmap = cv2.cvtColor(heatmap, cv2.COLOR_GRAY2BGR)
return image, heatmap
def flip(image, heatmap):
image = np.fliplr(image)
heatmap = np.fliplr(heatmap)
return image, heatmap
def distort(image):
def _convert(image, alpha=1, beta=0):
tmp = image.astype(float) * alpha + beta
tmp[tmp < 0] = 0
tmp[tmp > 255] = 255
image[:] = tmp
image = image.copy()
if random.randrange(2):
#brightness distortion
# if random.randrange(2):
# _convert(image, beta=random.uniform(-32, 32))
#contrast distortion
if random.randrange(2):
_convert(image, alpha=random.uniform(0.5, 1.5))
image = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
#saturation distortion
if random.randrange(2):
_convert(image[:, :, 1], alpha=random.uniform(0.5, 1.5))
#hue distortion
# if random.randrange(2):
# tmp = image[:, :, 0].astype(int) + random.randint(-18, 18)
# tmp %= 180
# image[:, :, 0] = tmp
image = cv2.cvtColor(image, cv2.COLOR_HSV2BGR)
else:
#brightness distortion
# if random.randrange(2):
# _convert(image, beta=random.uniform(-32, 32))
image = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
#saturation distortion
if random.randrange(2):
_convert(image[:, :, 1], alpha=random.uniform(0.5, 1.5))
#hue distortion
# if random.randrange(2):
# tmp = image[:, :, 0].astype(int) + random.randint(-18, 18)
# tmp %= 180
# image[:, :, 0] = tmp
image = cv2.cvtColor(image, cv2.COLOR_HSV2BGR)
#contrast distortion
if random.randrange(2):
_convert(image, alpha=random.uniform(0.5, 1.5))
return image
def add_gaussian_noise(X_imgs):
gaussian_noise_imgs = []
row, col, _ = X_imgs[0].shape
# Gaussian distribution parameters
mean = 0
var = 0.1
sigma = var ** 0.5
for X_img in X_imgs:
X_img = np.array(X_img, np.float32)
gaussian = np.random.random((row, col, 1)).astype(np.float32)
gaussian = np.concatenate((gaussian, gaussian, gaussian), axis = 2)
gaussian_img = cv2.addWeighted(X_img, 0.005, gaussian, 0.25, 0)
gaussian_noise_imgs.append(gaussian_img)
gaussian_noise_imgs = np.array(gaussian_noise_imgs, dtype = np.float32)
return gaussian_noise_imgs
def load_data(img_path):
if img_path.find("/rgb/") != -1:
gt_path = img_path.replace("/rgb/", "/masks/").replace("/image", "/label")
else:
gt_path = img_path.replace("/test_rgb/", "/test_masks/").replace("/image", "/label")
img = Image.open(img_path)
gt = Image.open(gt_path)
if img_path.find("/rgb/") != -1:
img, gt = crop(img, gt)
i = img
#chance = np.random.random()
# if chance < 0.0:
# img, gt = flip(img, gt)
# if chance < 0.2:
# img = distort(img)
if random.randrange(2):
value = random.randrange(5)
img[:,:,2] += value
img = np.clip(img, 0, 255)
# cv2.imshow( "brightness img",img)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
if random.randrange(2):
img = add_gaussian_noise([img])[0]
# cv2.imshow("img", np.array(i))
# cv2.imshow( "noise img",img)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# exit(1)
else:
img = np.array(img)
gt = np.array(gt)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
return img, gt
#add data aug functions
#return img