xcaliber-task/data_utils.py at master · brijml/xcaliber-task · GitHub

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
import numpy as np
import scipy.io as sio
import os
import matplotlib
# matplotlib.use("Agg")
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
from random import shuffle, randint
# from keras.utils.np_utils import to_categorical


def show(image):
	if len(image.shape) == 2:
		plt.imshow(image, cmap='Greys')
	else:
		plt.imshow(image)
	plt.show()
	return

def rand_shuffle(files):
	for i in range(20):
		shuffle(files)

	return files

def validation_split(gt_files, ratio):
	no_train = int(len(gt_files)*(1-ratio))
	train_gt = gt_files[0:no_train]
	val_gt = gt_files[no_train:]
	return train_gt, val_gt

def get_test_files(image_path, gt_path):
	test_files = []
	image_files = os.listdir(image_path)
	gt_files = os.listdir(gt_path)
	gt_mod = [i.split('.')[0] for i in gt_files]
	for i, file_ in enumerate(image_files):
		if file_.split('.')[0] not in gt_mod:
			test_files.append(file_)

	return test_files

def get_shuffled(images,labels):
	no = len(images)
	range_no = list(range(no))
	for i in range(20):
		shuffle(range_no)

	image_temp,labels_temp = [],[]
	for i in range_no:
		image_temp.append(images[i])
		labels_temp.append(labels[i])

	return image_temp,labels_temp

def get_batch(image, seg_mask, ratio=3, h=13, s=2):
	indices_pos = np.where(seg_mask == 1)
	indices_neg = np.where(seg_mask == 0)
	# print(len(indices_pos[0]))
	patches, labels = [], []
	m,n = image.shape[:2]
	count, length = 0, len(indices_pos[0])
	# for i,ind in enumerate(indices_pos[0]):
	while count < 100:
		rand_int = randint(0, length)-1
		x, y = indices_pos[0][rand_int], indices_pos[1][rand_int]#
		# x,y = ind, indices_pos[1][i]
		if x-h-1 < 0 or y-h-1 < 0 or x+h+1 > m or y+h+1>n:
			continue
		else:
			assert(image[x-h:x+h+1, y-h:y+h+1].shape == (27,27,3))
			patches.append(image[x-h:x+h+1, y-h:y+h+1])
			labels.append(seg_mask[x-s:x+s+1, y-s:y+s+1].flatten())
			count+=1

	no_negs = ratio*len(patches)
	count, length = 0, len(indices_neg[0])
	while count<no_negs:
		rand_int = randint(0, length)-1
		x, y = indices_neg[0][rand_int], indices_neg[1][rand_int]
		if x-h-1 < 0 or y-h-1 < 0 or x+h+1 > m or y+h+1>n:
			continue
		else:
			assert(image[x-h:x+h+1, y-h:y+h+1].shape == (27,27,3))
			patches.append(image[x-h:x+h+1, y-h:y+h+1])
			labels.append(seg_mask[x-s:x+s+1, y-s:y+s+1].flatten())
			count+=1

	patches, labels = get_shuffled(patches, labels)
	return np.array(patches), np.array(labels)

def load_gen(image_path, gt_path, gt_files, batch_size, n_classes):
	image_files = os.listdir(image_path)
	# gt_files = os.listdir(gt_path)
	gt_files = rand_shuffle(gt_files)
	for i in range(0,len(gt_files),batch_size):
		gt_files_batch = gt_files[i:i+batch_size]
		images, labels = [], []
		for gt_file in gt_files_batch:
			image_file = gt_file.split('.')[0]+'.jpg'
			img = mpimg.imread(os.path.join(image_path, image_file))
			mat_contents = sio.loadmat(os.path.join(gt_path, gt_file))
			seg_mask = mat_contents['groundTruth'][0][0][0]
			seg_mask[seg_mask == 1] = 0
			seg_mask[seg_mask > 1] = 1
			images.append(img)
			labels.append(seg_mask)

		categorical_labels = to_categorical(np.array(labels), num_classes=n_classes)
		yield np.array(images), categorical_labels

def load_gen_v2(image_path, gt_path, gt_files):#, batch_size, n_classes):
	image_files = os.listdir(image_path)
	# gt_files = os.listdir(gt_path)
	gt_files = rand_shuffle(gt_files)
	for gt_file in gt_files:
		image_file = gt_file.split('.')[0]+'.jpg'
		img = mpimg.imread(os.path.join(image_path, image_file))
		mat_contents = sio.loadmat(os.path.join(gt_path, gt_file))
		seg_mask = mat_contents['groundTruth'][0][0][0]
		seg_mask[seg_mask == 1] = 0
		seg_mask[seg_mask > 1] = 1
		yield get_batch(img, seg_mask)


if __name__ == '__main__':
	base_path = '/home/brijml/Desktop/CrackForest-dataset-master'
	image_path = os.path.join(base_path, 'image')
	gt_path = os.path.join(base_path, 'groundTruth')
	gt_files = os.listdir(gt_path)
	gen_object = load_gen_v2(image_path, gt_path, gt_files)#, 10, 2)
	for batch_train in gen_object:
		images = batch_train[0]
		labels = batch_train[1]
		print(images.shape)
		# for i,image in enumerate(images):
			# show(image)
			# show(labels[i][:,:,0])
			# show(labels[i][:,:,1])
			# print(labels[i])