-
Notifications
You must be signed in to change notification settings - Fork 0
/
lstmincresatten1ntu.py
348 lines (254 loc) · 11.4 KB
/
lstmincresatten1ntu.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
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
#!/usr/bin/env python
# coding: utf-8
# In[ ]:
#!/usr/bin/env python
# coding: utf-8
# In[ ]:
#!/usr/bin/env python
# coding: utf-8
# In[ ]:
from __future__ import division,print_function
import math, os, json, sys, re
# import cPickle as pickle
from glob import glob
# from fast.ai.imports import *
import PIL
from PIL import Image
import numpy as np
from numpy.random import random, permutation, randn, normal, uniform, choice
from numpy import newaxis
import scipy
from scipy import misc, ndimage
from scipy.ndimage.interpolation import zoom
# from scipy.ndimage import imread
#from sklearn.metrics import confusion_matrix
# import bcolz
from sklearn.preprocessing import OneHotEncoder
from sklearn.manifold import TSNE
#from IPython.lib.display import FileLink
# import theano
# from theano import shared, tensor as T
# from theano.tensor.nnet import conv2d, nnet
# from theano.tensor.signal import pool
import keras
from keras import backend as K
from keras.utils.data_utils import get_file
from keras.utils import np_utils
from keras.utils.np_utils import to_categorical
from keras.models import Sequential, Model,load_model
from keras.metrics import top_k_categorical_accuracy
from keras.layers.core import Flatten, Dense, Dropout, Lambda
from keras.layers.convolutional import *
from keras.preprocessing import image, sequence
from keras.optimizers import SGD, RMSprop
import keras.callbacks as kcallbacks
from keras.applications import VGG16,ResNet50
import numpy as np
from scipy import ndimage
from keras.preprocessing.image import ImageDataGenerator, array_to_img, img_to_array, load_img
from keras.applications.inception_resnet_v2 import InceptionResNetV2, preprocess_input
from keras.layers import Input, Conv2D, Activation, BatchNormalization, GlobalAveragePooling2D, Dense, Dropout,Add,Multiply,LSTM,Reshape,LeakyReLU
from keras.layers.merge import add
from keras.layers.merge import concatenate
class TestCallback(keras.callbacks.Callback):
def __init__(self, test_data):
self.test_data = test_data
self.xx=0
def on_epoch_end(self, epoch, logs={}):
x, y = self.test_data
loss, acc,top_3_accuracy,top_5_accuracy = self.model.evaluate(x, y, verbose=0)
print('\nTesting loss: {}, acc: {},top_3_accuracy: {},top_5_accuracy: {}\n'.format(loss,acc,top_3_accuracy,top_5_accuracy))
if(top_5_accuracy > self.xx):
model1_pred = model.predict(dataset1,batch_size=4)
classes = np.argmax(model1_pred,axis=1)
np.save('/home/ws2/Documents/msrpredarray/pred_msr_fuskdata.npy', model1_pred)
print(classes)
self.xx= top_5_accuracy
IMAGE_SIZE = [224, 224] # we will keep the image size as (64,64). You can increase the size for better results.
# loading the weights of VGG16 without the top layer. These weights are trained on Imagenet dataset.
#vgg = VGG16(input_shape = IMAGE_SIZE + [3], weights = 'imagenet', include_top = False) # input_shape = (64,64,3) as required by VGG
#base_model = InceptionResNetV2(input_shape = IMAGE_SIZE + [3],include_top=False )
#base_model = ResNet50(input_shape = IMAGE_SIZE + [3],include_top=False )
# def save_array(fname, arr):
# c=bcolz.carray(arr, rootdir=fname, mode='w')
# c.flush()
# def load_array(fname):
# return bcolz.open(fname)[:]
#path = "/home/ws2/Documents/cropdata/"
#path ="/home/ws2/Documents/finaldata1/seg4/train/head_spinal/"
#path ="/home/ws2/Documents/finaldata1/seg4/train/left_hand/"
#path ="/home/ws2/Documents/finaldata1/seg4/train/left_leg/"
#path ="/home/ws2/Documents/finaldata1/seg4/train/right_hand/"
#path ="/home/ws2/Documents/finaldata1/seg4/train/right_leg/"
#path ="/home/ws2/Documents/traintest_dataparts/seg1/train/head_spinal/"
path ="/home/ws2/Documents/ttfidmsr_fusk/train/"
#path ="/home/ws2/Documents/cropdata/train"
#path ="/home/ws2/Documents/traintest_dataparts/seg1/train/left_hand/"
#path ="/home/ws2/Documents/traintest_dataparts/seg1/train/left_leg/"
#path ="/home/ws2/Documents/traintest_dataparts/seg1/train/right_hand/"
#path ="/home/ws2/Documents/traintest_dataparts/seg1/train/right_leg/"
#path ="/home/ws2/Documents/kinectdata/seg2/"
#path ="/home/ws2/Documents/kinectdata/seg3/"
#path ="/home/ws2/Documents/kinectdata/seg4/"
# path = "data/dogscats/"
model_path = path + 'models/'
if not os.path.exists(model_path): os.mkdir(model_path)
from keras.layers import Dense, Conv2D, BatchNormalization, Activation
from keras.layers import AveragePooling2D, Input, Flatten
from keras.optimizers import Adam,SGD
from keras.models import Model
input_tensor = Input((224, 224, 3))
def AttentionBlock(x,shortcut,i_filters):
g1 = Conv2D(i_filters,kernel_size = 1)(shortcut)
g1 = BatchNormalization()(g1)
x1 = Conv2D(i_filters,kernel_size = 1)(x)
x1 = BatchNormalization()(x1)
g1_x1 = Add()([g1,x1])
psi = Activation('relu')(g1_x1)
psi = Conv2D(1,kernel_size = 1)(psi)
psi = BatchNormalization()(psi)
psi = Activation('sigmoid')(psi)
x = Multiply()([x,psi])
return x
shortcut=MaxPooling2D(pool_size=(3,3), strides=(1,1), padding='same')(input_tensor)
p=AttentionBlock(input_tensor,shortcut,16)
x1=Conv2D(64,kernel_size=(1,1),padding='same',strides=(1,1))(input_tensor)
x1=BatchNormalization()(x1)
#x1=MaxPooling2D(pool_size=(3,3), strides=(1,1))(x1)
x2=Conv2D(32,kernel_size=(1,1),padding='same',strides=(1,1))(input_tensor)
x2=BatchNormalization()(x2)
x2=Activation('relu')(x2)
#x2=LeakyReLU(alpha=0.1)(x2)
x2=Conv2D(64,kernel_size=(3,3),padding='same',strides=(1,1))(x2)
x2=BatchNormalization()(x2)
#x2=MaxPooling2D(pool_size=(3,3), strides=(1,1))(x2)
x3=Conv2D(32,kernel_size=(1,1),padding='same',strides=(1,1))(input_tensor)
x3=BatchNormalization()(x3)
x3=Activation('relu')(x3)
#x3=LeakyReLU(alpha=0.1)(x3)
x3=Conv2D(64,kernel_size=(3,3),padding='same',strides=(1,1))(x3)
x3=BatchNormalization()(x3)
#x3=MaxPooling2D(pool_size=(3,3), strides=(1,1))(x3)
x4=MaxPooling2D(pool_size=(3,3), strides=(1,1), padding='same')(input_tensor)
x4=Conv2D(32,kernel_size=(1,1),padding='same',strides=(1,1))(x4)
x4=BatchNormalization()(x4)
#x4=MaxPooling2D(pool_size=(3,3), strides=(1,1))(x4)
layer_out1 = concatenate([x1, x2, x3,x4], axis=-1)
x5=Conv2D(224,kernel_size=(1,1),padding='same',strides=(1,1))(p)
#x5=MaxPooling2D(pool_size=(3,3), strides=(1,1))(x5)
h = add([layer_out1, x5])
h=Activation('relu')(h)
#h=LeakyReLU(alpha=0.1)(h)
#x6=Conv2D(32,kernel_size=(1,1),strides=(1,1))(h)
#x6=BatchNormalization()(x6)
#x6=MaxPooling2D(pool_size=(3,3), strides=(1,1))(x6)
#x10=Conv2D(32,kernel_size=(3,3),strides=(1,1))(p)
#x10=MaxPooling2D(pool_size=(3,3), strides=(1,1))(x10)
#h = add([x6, x10])
#h=Activation('relu')(h)
h = GlobalAveragePooling2D()(h)
#print(h.output_shape)
h=Reshape((1,-1))(h)
h = BatchNormalization()(h)
h=LSTM(256)(h)
h=Reshape((1,-1))(h)
h=LSTM(128)(h)
#h=Dense(128)(h)
#h=Activation('relu')(h)
#h=LeakyReLU(alpha=0.1)(h)
#h = BatchNormalization()(h)
output = Dense(8, activation='softmax')(h)
model = Model(inputs=input_tensor, outputs=output)
print(model.summary())
# In[14]:
gen=image.ImageDataGenerator()
folder = path+'valid'
categories=['a02','a03','a05','a06','a10','a13','a18','a20']
#categories=['carry','clapHands','pickUp','pull','push','sitDown','standUp','throw','walk','waveHands']
training_data= []
#onlyfiles = [f for f in os.listdir(folder) if os.path.isfile(os.path.join(folder, f))]
from scipy import ndimage
from keras.preprocessing.image import ImageDataGenerator, array_to_img, img_to_array, load_img
train_files = []
y_train = []
i=0
def dt(folder1,cat):
dataset=[]
j = 0
y1=[]
for category in categories:
for file in os.listdir(folder1+'/'+str(category)):
#print(file)
img = load_img(folder1+'/'+str(category)+'/'+file)
img.thumbnail((224,224))
img = img.resize((224,224), Image.ANTIALIAS)
# Convert to Numpy Array
x = img_to_array(img)
x = x.reshape((224,224,3))
dataset.append(x)
#print(categories.index(category))
y1.append(categories.index(category))
dataset=np.array(dataset)
values = np.array(y1)
#print("All images to array!")
return dataset,values
#print(y)
dataset,values=dt('/home/ws2/Documents/ttfidmsr_fusk/train',categories)
#print(dataset.shape)
#print(values.shape)
dataset1,values1=dt('/home/ws2/Documents/ttfidmsr_fusk/valid',categories)
# In[ ]:
#y_labels=onehot(y)
from sklearn.preprocessing import LabelEncoder
from sklearn.preprocessing import OneHotEncoder
#print(values)
# integer encode
label_encoder = LabelEncoder()
integer_encoded = label_encoder.fit_transform(values)
#print(integer_encoded)
onehot_encoder = OneHotEncoder(sparse=False)
integer_encoded = integer_encoded.reshape(len(integer_encoded), 1)
onehot_encoded = onehot_encoder.fit_transform(integer_encoded)
integer_encoded1 = label_encoder.fit_transform(values1)
#print(integer_encoded)
onehot_encoder = OneHotEncoder(sparse=False)
integer_encoded1 = integer_encoded1.reshape(len(integer_encoded1), 1)
onehot_encoded1 = onehot_encoder.fit_transform(integer_encoded1)
def top_5_accuracy(y_true, y_pred):
return top_k_categorical_accuracy(y_true,y_pred,k=5)
def top_3_accuracy(y_true, y_pred):
return top_k_categorical_accuracy(y_true,y_pred,k=3)
# In[ ]:
#nb_epoch=50
#p=TestCallback((dataset, onehot_encoded))
best_weights_filepath = model_path+'aril_best_weights12.hdf5'
#earlyStopping=kcallbacks.EarlyStopping(monitor='val_loss', patience=10, verbose=1, mode='auto')
saveBestModel = kcallbacks.ModelCheckpoint(best_weights_filepath, monitor='val_top_5_accuracy', verbose=1, save_best_only=True, mode='auto')
#,fill_mode='reflect'
#callbacks=[saveBestModel,p]
aug = ImageDataGenerator(rotation_range=60, zoom_range=0.2,fill_mode='nearest',
width_shift_range=0.1, height_shift_range=0.1, shear_range=0.01,
horizontal_flip=True)
def fit_model(model, dataset1,onehot_encoded1,dataset,onehot_encoded):
history = model.fit_generator(aug.flow(dataset,onehot_encoded, batch_size=64),
validation_data=(dataset1,onehot_encoded1), steps_per_epoch=len(dataset) //64,
epochs=1000,callbacks=[saveBestModel,TestCallback((dataset1, onehot_encoded1))])
print(history.history.keys())
# summarize history for accuracy
acc=(history.history['top_5_accuracy'])
np.save('/home/ws2/Documents/newmodelsarray_msrfusk/acc_seg1_msrfusk.npy', acc)
acc1=(history.history['acc'])
np.save('/home/ws2/Documents/newmodelsarray_msrfusk/acc1_seg1_msrfusk.npy', acc1)
acc2=(history.history['top_3_accuracy'])
np.save('/home/ws2/Documents/newmodelsarray_msrfusk/acc2_seg1_msrfusk.npy', acc2)
val_acc=(history.history['val_top_5_accuracy'])
np.save('/home/ws2/Documents/newmodelsarray_msrfusk/val_acc_seg1_msrfusk.npy', val_acc)
loss=(history.history['loss'])
np.save('/home/ws2/Documents/newmodelsarray_msrfusk/loss_seg1_msrfusk.npy', loss)
val_loss=(history.history['val_loss'])
np.save('/home/ws2/Documents/newmodelsarray_msrfusk/val_loss_seg1_msrfusk.npy', val_loss)
opt = Adam(lr=0.001,decay=1e-6)
#model.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])
model.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['acc',top_3_accuracy,top_5_accuracy])
fit_model(model, dataset1,onehot_encoded1,dataset,onehot_encoded)
#model.save_weights(model_path+'after_test_weights.hdf5')