Convolutional-Encoder-Decoder-for-Hand-Segmentation/encoder-decoder-train.py at master · sharnam19/Convolutional-Encoder-Decoder-for-Hand-Segmentation · 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
from keras.models import Sequential
from keras.layers.convolutional import Convolution2D
from keras.layers import Activation
from keras.layers import MaxPooling2D,UpSampling2D
from keras.layers import Dropout,Dense,Flatten,BatchNormalization
from keras.optimizers import *
from keras.models import load_model
from keras import regularizers
import os
import cv2
import numpy as np
import sys
angles = range(-2,3)
shifts = [[0,0],[0,1],[1,0],[1,1],[0,2],[2,0],[1,2],[2,1],[2,2],
                [0,-1],[-1,0],[-1,-1],[0,-2],[-2,0],[-1,-2],[-2,-1],[-2,-2],
                [1,-1],[1,-2],[2,-1],[2,-2],
                [-1,1],[-1,2],[-2,1],[-2,2]]
multiplier = len(angles)*len(shifts)
X_train=np.zeros((100*multiplier,128,128,3))
y_train=np.zeros((100*multiplier,128,128,3))

path_x = 'Data/X/'
path_y = 'Data/Y2/'
total = 0
#for pos in range(len(path_x)):
for img in sorted(os.listdir(path_x))[:-5]:
    originalIm = cv2.imread(path_x+img)
    segmentedIm = cv2.imread(path_y+img)

    for angle in angles:
        for shift in shifts :

            M = cv2.getRotationMatrix2D((128/2,128/2),angle,1)
            shiftM = np.float32([[1,0,shift[0]],[0,1,shift[1]]])
            rotatedIm = cv2.warpAffine(originalIm,M,(128,128))
            rotatedSegmentedIm = cv2.warpAffine(segmentedIm,M,(128,128))
            rotatedShiftedIm = cv2.warpAffine(rotatedIm,shiftM,(128,128))
            rotatedSegmentedShiftedIm = cv2.warpAffine(rotatedSegmentedIm,shiftM,(128,128))
            X_train[total]=rotatedShiftedIm
            y_train[total]=rotatedSegmentedShiftedIm
            total+=1

X_test = np.zeros((5,128,128,3))
tests = ["A-train0101.jpg","A-train0102.jpg","A-train0103.jpg","A-train0104.jpg","A-train0105.jpg"]
for pos in range(len(tests)):
    X_test[pos] = cv2.imread(path_x+tests[pos])
X_train-=128.0
X_train/=128.0
y_train-=128.0
y_train/=128.0
X_test-=128.0
X_test/=128.0
#
# meen = np.mean(X_train,axis=(0,1,2))
# std = np.std(X_train,axis=(0,1,2))
# X_train-=meen
# X_train/=std
#
# #y_train-=meen
# y_train/=255
#

adam = Adam(lr=0.01, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
clf = Sequential()

clf.add(Convolution2D(filters=64,kernel_size=(3,3),input_shape=(128,128,3), padding='same'))
clf.add(BatchNormalization())
clf.add(Activation('relu'))
clf.add(MaxPooling2D(pool_size=(2,2)))

clf.add(Convolution2D(filters=128,kernel_size=(3,3),padding='same'))
clf.add(BatchNormalization())
clf.add(Activation('relu'))
clf.add(MaxPooling2D(pool_size=(2,2)))

clf.add(Convolution2D(filters=256,kernel_size=(3,3), padding='same'))
clf.add(BatchNormalization())
clf.add(Activation('relu'))
clf.add(MaxPooling2D(pool_size=(2,2)))

clf.add(Convolution2D(filters=512,kernel_size=(3,3), padding='same'))
clf.add(BatchNormalization())
clf.add(Activation('relu'))

clf.add(Convolution2D(filters=512,kernel_size=(3,3), padding='same'))
clf.add(BatchNormalization())
clf.add(Activation('relu'))

clf.add(UpSampling2D((2,2)))
clf.add(Convolution2D(filters=256,kernel_size=(3,3), padding='same'))
clf.add(BatchNormalization())
clf.add(Activation('relu'))

clf.add(UpSampling2D((2,2)))
clf.add(Convolution2D(filters=128,kernel_size=(3,3), padding='same'))
clf.add(BatchNormalization())
clf.add(Activation('relu'))

clf.add(UpSampling2D((2,2)))
clf.add(Convolution2D(filters=64,kernel_size=(3,3), padding='same'))
clf.add(BatchNormalization())
clf.add(Activation('relu'))

clf.add(Convolution2D(3, (3, 3), padding='same'))
clf.add(Activation('tanh'))

clf.compile(optimizer=adam,loss='mse',metrics=['mae'])
clf.fit(X_train,y_train,batch_size=30, epochs=10,validation_split=0.2)
sys.stdout.flush()
y_out = clf.predict(X_test)
y_out*=128.0
y_out+=128.0
clf.save('model-10.h5')
for y in range(y_out.shape[0]):
    cv2.imwrite('y'+str(y)+'.jpg',y_out[y])