-
Notifications
You must be signed in to change notification settings - Fork 0
/
hsv_rf.py
205 lines (153 loc) · 5.6 KB
/
hsv_rf.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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
from sklearn.ensemble import RandomForestClassifier
import glob
from skimage import io,img_as_uint, img_as_float
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score,mean_squared_error, jaccard_similarity_score, confusion_matrix
from sklearn.externals import joblib
from skimage.feature import hog
from skimage.color import rgb2gray
import matplotlib.pyplot as plt
from skimage import data, exposure, filters
from skimage import io; io.use_plugin('matplotlib')
from skimage.color import rgb2hsv
import matplotlib.cm as cm
# return a list of images
def read_data(imgDir, maskDir):
images = glob.glob(imgDir + "*.jpg")
masks = glob.glob(maskDir + "*.png")
train = []
label = []
for im in images:
n = im.replace(imgDir,'')
n = n.replace('.jpg','')
train.append(n)
for mask in masks:
n = mask.replace(maskDir,'')
n = n.replace('.png','')
label.append(n)
if ( set(label) == set(train)):
print("each image has mask")
x = []
y = []
original = []
for i in range(len(train)):
name = imgDir + str(train[i]) + '.jpg'
image = io.imread(name)
original.append(image)
#fd, hog_image = hog( image, orientations=8, pixels_per_cell=(2, 2),
# cells_per_block=(1, 1), visualize=True, multichannel=True)
hsv = filters.gaussian(image,sigma=3)
hsv = rgb2hsv(hsv)
#m = np.concatenate([image.ravel(),hsv.ravel()])
#m = np.concatenate([hsv.ravel(),fd])
#m = hsv.ravel()
x.append(hsv.ravel())
name = maskDir + str(train[i]) + '.png'
mask = img_as_uint(io.imread(name))
print(np.mean(mask))
print(np.unique(mask))
mask = np.where(mask > 0.5, 1, 0) # convert to binary target pixel values
print(np.mean(mask))
print(np.unique(mask))
print()
y.append(mask.ravel())
return x,y,original
def train_model(train_x, train_y, test_x, test_y):
print("start train model")
# Create random forest classifier instance
# trained_model = RandomForestClassifier(verbose=1, n_estimators=1,warm_start=True,n_jobs=-1)
# batch_size = 4
# split = len(train_x) / batch_size
# print("Split data to: " + str(split))
# step = 0
# while (step < split):
# print("Step number: " + str(step))
# trained_model.n_estimators = trained_model.n_estimators + 4
# trained_model.fit(train_x[step*batch_size : (step+1)*batch_size], train_y[step*batch_size : (step+1)*batch_size])
# print("length of batch: " + str(len(train_x[step*batch_size : (step+1)*batch_size])))
# step += 1
trained_model = RandomForestClassifier(verbose=1, n_estimators=4,warm_start=True, n_jobs=-1)
batch_size = 2
split = len(train_x) / batch_size + 1
print("Split data to :" + str(split-1))
step = 1
while (step < split):
print("Step number: " + str(step - 1))
print("length of batch: " + str(len(train_x[batch_size*(step-1):step*batch_size])))
trained_model.fit(train_x[batch_size*(step-1):step*batch_size], train_y[batch_size*(step-1):step*batch_size])
trained_model.n_estimators = trained_model.n_estimators + 4
step += 1
trained_model.n_estimators = trained_model.n_estimators - 4
print("Trained model :: " +str(trained_model) )
predictions = trained_model.predict(test_x[0])
pred = []
for p in predictions:
pred.append(np.array(p,dtype=int))
print(predictions)
print(np.mean(predictions[0]))
print(np.mean(test_y[0]))
miou = 0
for p,y in zip(pred,test_y):
miou = miou + cal_miou(p,y)
miou = miou / len(pred)
print("mIOU :: " + str(miou))
s = joblib.dump(trained_model, 'model.pkl',compress=9)
print("model saved")
def pred_model(test_x, test_y,original_images):
print("start load model")
trained_model = joblib.load('model_hsv_rf_240.pkl')
print("model loaded")
predictions = trained_model.predict(test_x)
pred = []
for p in predictions:
pred.append(np.array(p,dtype=int))
miou = 0
for p,y in zip(pred,test_y):
miou = miou + cal_miou(p,y)
miou = miou / len(pred)
print("mIOU :: " + str(miou))
print(predictions[0][0])
for i in range(len(predictions)):
print(i)
io.imsave("./out/" + str(i) + "_pred.tif", img_as_float(predictions[i].reshape(256,256,1)))
io.imsave("./out/" + str(i) + "_img.tif", img_as_uint(original_images[i]) )
#plt.imsave("./out/" + str(i) + "_mask.png", np.array(test_y[i].reshape(256,256,1)), cmap=cm.gray)
# plt.imsave("./out/" + str(i) + "_img.png", np.array(test_x[i].reshape(256,256,3)), cmap=cm.gray)
# background = Image.open("./out/" + str(i) + "_pred.png")
# overlay = Image.open("./out/" + str(i) + "_img.png")
# background = background.convert("RGBA")
# overlay = overlay.convert("RGBA")
# new_img = Image.blend(background, overlay, 0.5)
# new_img.save("./out/" + str(i) + "_new.png","PNG")
def cal_miou(pred,val):
tn, fp, fn, tp = confusion_matrix(val,pred).ravel()
miou = tp / float(tp + fp + fn)
print(miou)
return miou
## load data
imgDir = './data_small/train_images/'
maskDir = './data_small/train_masks/'
imgValDir = './data_full/validate_images/'
maskValDir = './data_full/validate_masks/'
train_x,train_y,_ = read_data(imgDir, maskDir)
test_x, test_y,original_images = read_data(imgValDir, maskValDir)
print("train set: " + str(len(train_x)))
print("test set: " + str(len(test_x)))
print(len(train_x[0]))
print(len(train_x))
train_count = 8
print("Training on: " + str(train_count))
print(train_x[0:train_count])
print(train_y[0:train_count])
print(test_x)
print(test_y)
print
batch_size = 4
step = 1
print(train_x[batch_size*(step-1):step*batch_size])
print(train_y[batch_size*(step-1):step*batch_size])
step = 2
print(train_x[batch_size*(step-1):step*batch_size])
#train_model(train_x[0:train_count],train_y[0:train_count],test_x, test_y)
pred_model(test_x, test_y,original_images)