-
Notifications
You must be signed in to change notification settings - Fork 16
/
CRF_FineTune.py
176 lines (139 loc) · 6.43 KB
/
CRF_FineTune.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
import argparse
import cv2 as cv
import os
import pickle
import time
import numpy as np
import pydensecrf.densecrf as dcrf
from keras.models import load_model
from pydensecrf.utils import create_pairwise_bilateral, create_pairwise_gaussian
import .acc_util as au
from .acc_ass import accuracy_assessment
from .net_util import weight_binary_cross_entropy
from keras.layers import Lambda
import keras.backend as K
from .seg_model.MyModel.SiameseInception_Keras import SiameseInception
parser = argparse.ArgumentParser()
parser.add_argument('--data_path', default='./data/ACD', help='data path')
parser.add_argument('--data_set_name', default='Szada', help='dataset name')
# basic params
FLAGS = parser.parse_args()
DATA_PATH = FLAGS.data_path
DATA_SET_NAME = FLAGS.data_set_name
def fine_tune_result():
path = os.path.join(DATA_PATH, DATA_SET_NAME)
test_X, test_Y, test_label = load_test_data(path=path)
test_X = np.array(test_X) / 255.
test_Y = np.array(test_Y) / 255.
test_label = np.array(test_label) / 255.
test_label = np.reshape(test_label, (test_label.shape[0], test_label.shape[1], test_label.shape[2]))
# test = np.concatenate([test_X, test_Y], axis=-1)
# MS_model = load_model("393_model.h5",
# custom_objects={
# 'weight_binary_cross_entropy': weight_binary_cross_entropy,
# 'Recall': au.Recall,
# 'Precision': au.Precision,
# 'F1_score': au.F1_score
# })
#
Network = SiameseInception()
MS_model = Network.get_model(input_size=[None, None, 3])
MS_model.load_weights("./model_param/ACD/Szada/DSMSFCN/27_model.h5", by_name=True)
MS_model.compile(optimizer='Adam', loss=weight_binary_cross_entropy,
metrics=['accuracy', au.Recall, au.Precision, au.F1_score])
loss, acc, sen, spe, F1 = MS_model.evaluate(x=[test_X, test_Y], y=test_label, batch_size=1)
tic = time.time()
for i in range(0, 1):
change_prob = MS_model.predict([test_X, test_Y])
toc = time.time()
change_prob_2 = np.array(255 * change_prob, dtype=np.uint8)
cv.imwrite('CIM.bmp', change_prob_2[0])
print('network time: ', (toc - tic))
diff = np.abs(test_X - test_Y)
diff = 255 * (diff - np.min(diff)) / (np.max(diff) - np.min(diff))
image = np.squeeze(diff, axis=0)
test_label = np.squeeze(test_label, axis=0)
binary_change_map = np.copy(np.reshape(change_prob, (change_prob.shape[1], change_prob.shape[2])))
idx_1 = binary_change_map > 0.5
idx_2 = binary_change_map <= 0.5
binary_change_map[idx_1] = 255
binary_change_map[idx_2] = 0
conf_mat, overall_acc, kappa = accuracy_assessment(
gt_changed=np.reshape(255 * test_label, (test_label.shape[0], test_label.shape[1])),
gt_unchanged=np.reshape(255. - 255 * test_label, (test_label.shape[0], test_label.shape[1])),
changed_map=binary_change_map)
print(conf_mat)
info = 'loss is %.4f, sen is %.4f, spe is %.4f, F1 is %.4f, acc is %.4f, kappa is %.4f, ' % (
loss, sen, spe, F1, overall_acc, kappa)
print(info)
# change_prob = np.expand_dims(change_prob, axis=-1)
unchange_prob = 1. - change_prob
softmax_result = np.concatenate([change_prob, unchange_prob], axis=0)
# softmax_result = np.transpose(softmax_result, axes=[2, 0, 1])
# unary potential
# unary = softmax_to_unary(softmax_result)
# unary = np.ascontiguousarray(unary) # (2, n)
unary = -np.log(softmax_result)
unary = unary.reshape((2, -1))
unary = np.ascontiguousarray(unary)
d = dcrf.DenseCRF(binary_change_map.shape[1] * binary_change_map.shape[0], 2)
d.setUnaryEnergy(unary)
# This potential penalizes small pieces of segmentation that are
# spatially isolated -- enforces more spatially consistent segmentations
feats = create_pairwise_gaussian(sdims=(3, 3), shape=binary_change_map.shape[:2])
d.addPairwiseEnergy(feats, compat=3,
kernel=dcrf.DIAG_KERNEL,
normalization=dcrf.NORMALIZE_SYMMETRIC)
# This creates the color-dependent features --
# because the segmentation that we get from CNN are too coarse
# and we can use local color features to refine them
feats = create_pairwise_bilateral(sdims=(7, 7), schan=(10, 10, 5),
img=image, chdim=2)
d.addPairwiseEnergy(feats, compat=4,
kernel=dcrf.DIAG_KERNEL,
normalization=dcrf.NORMALIZE_SYMMETRIC)
tic = time.time()
for i in range(0, 1):
Q = d.inference(20)
toc = time.time()
print('FC-CRF time: ', (toc - tic))
res = np.argmax(Q, axis=0).reshape((binary_change_map.shape[0], binary_change_map.shape[1]))
recall = Recall(y_true=test_label, y_pred=1. - res)
pre = Precision(y_true=test_label, y_pred=1. - res)
f1 = F1_score(y_true=test_label, y_pred=1. - res)
print(recall, pre, f1)
idx_1 = res == 0
idx_2 = res == 1
res[idx_1] = 255
res[idx_2] = 0
cv.imwrite('another_result.bmp', res)
conf_mat, overall_acc, kappa = accuracy_assessment(
gt_changed=np.reshape(255 * test_label, (test_label.shape[0], test_label.shape[1])),
gt_unchanged=np.reshape(255. - 255 * test_label, (test_label.shape[0], test_label.shape[1])),
changed_map=res)
print(conf_mat)
print(overall_acc, kappa)
def load_test_data(path):
with open(os.path.join(path, 'test_sample_1.pickle'), 'rb') as file:
test_X = pickle.load(file)
with open(os.path.join(path, 'test_sample_2.pickle'), 'rb') as file:
test_Y = pickle.load(file)
with open(os.path.join(path, 'test_label.pickle'), 'rb') as file:
test_label = pickle.load(file)
return test_X, test_Y, test_label
def Abs_layer(tensor):
return Lambda(K.abs)(tensor)
def Recall(y_true, y_pred):
true_positives = np.sum(np.round(np.clip(y_true * y_pred, 0, 1)))
possible_positives = np.sum(np.round(np.clip(y_true, 0, 1)))
return true_positives / (possible_positives + 1e-8)
def Precision(y_true, y_pred):
true_positives = np.sum(np.round(np.clip(y_true * y_pred, 0, 1)))
possible_negatives = np.sum(np.round(np.clip(y_pred, 0, 1)))
return true_positives / (possible_negatives + 1e-8)
def F1_score(y_true, y_pred):
R = Recall(y_true, y_pred)
P = Precision(y_true, y_pred)
return 2 * P * R / (R + P)
if __name__ == '__main__':
fine_tune_result()