forked from stnamin/Deep-Phenotyping
-
Notifications
You must be signed in to change notification settings - Fork 0
/
cnn_lstm_perframe.py
356 lines (298 loc) · 12.3 KB
/
cnn_lstm_perframe.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
348
349
350
351
352
353
354
355
##### This code is developed to extract deep features using cnn, train and test lstm (label each frame of sequence) ####
from __future__ import print_function
from keras.models import Sequential,Model
from keras.layers import LSTM, SimpleRNN
from keras.layers import Convolution2D, MaxPooling2D, ZeroPadding2D, TimeDistributed
from keras.optimizers import SGD, rmsprop
#from keras.utils.data_utils import get_file
#from sklearn.metrics import confusion_matrix
from keras.layers import Flatten, Dense, Dropout, Activation, Reshape, Permute, Input, merge
from keras.layers.convolutional import Convolution2D, MaxPooling2D, ZeroPadding2D
from keras.optimizers import SGD
import cv2, numpy as np
from keras.datasets import cifar10
from keras.utils import np_utils
#from keras.optimizers import SGD
from convnetskeras.convnets import preprocess_image_batch, convnet
from keras.callbacks import ModelCheckpoint
import random
import csv
from sklearn.metrics import confusion_matrix
np.set_printoptions(threshold=np.nan)
#########################################################
#### Reading the images
#########################################################
INPUT_LEN = 22
INPUT_DIM = 4096
OUTPUT_LEN = 4
batch_size = 32
nb_classes = 4
data_augmentation = False
# input image dimensions
img_rows, img_cols = 320, 320
# the images are RGB
img_channels = 3
data=np.load('BVZ0072_BVZ0073_97pots_with_area.npz')
print('hey')
TOT_0=data['TOT_0_seg']
TOT_90=data['TOT_90_seg']
TOT_180=data['TOT_180_seg']
TOT_270=data['TOT_270_seg']
GL=data['Plant_label']
y1=data['Plant_label_single']
Tray=data['Tray_label_single']
BVZ=data['BVZ_label_single']
TOT_0=np.transpose(TOT_0,(0,3,1,2))
TOT_90=np.transpose(TOT_90,(0,3,1,2))
TOT_180=np.transpose(TOT_180,(0,3,1,2))
TOT_270=np.transpose(TOT_270,(0,3,1,2))
#TOT_0 = TOT_0.astype('float32')
#TOT_90 = TOT_90.astype('float32')
#TOT_180 = TOT_180.astype('float32')
#TOT_270 = TOT_270.astype('float32')
#X_train /= 255
#X_test /= 255
m=[103.939, 116.779, 123.68]
TOT_0_r = np.zeros((TOT_0.shape[0],3,227,227),dtype=np.uint8)
for i in range(TOT_0.shape[0]):
temp = TOT_0[i,0,:,:]
temp2 = cv2.resize(temp,(227,227))
TOT_0_r[i,2,:,:] = temp2-m[0]
temp = TOT_0[i,1,:,:]
temp2 = cv2.resize(temp,(227,227))
TOT_0_r[i,1,:,:] = temp2-m[1]
temp = TOT_0[i,2,:,:]
temp2 = cv2.resize(temp,(227,227))
TOT_0_r[i,0,:,:] = temp2-m[2]
TOT_0_r[i,:,:,:] = np.expand_dims(TOT_0_r[i,:,:,:], axis=0)
print(TOT_0_r.shape[2])
TOT_90_r = np.zeros((TOT_90.shape[0],3,227,227),dtype=np.uint8)
for i in range(TOT_90.shape[0]):
temp = TOT_90[i,0,:,:]
temp2 = cv2.resize(temp,(227,227))
TOT_90_r[i,2,:,:] = temp2-m[0]
temp = TOT_90[i,1,:,:]
temp2 = cv2.resize(temp,(227,227))
TOT_90_r[i,1,:,:] = temp2-m[1]
temp = TOT_90[i,2,:,:]
temp2 = cv2.resize(temp,(227,227))
TOT_90_r[i,0,:,:] = temp2-m[2]
TOT_90_r[i,:,:,:] = np.expand_dims(TOT_90_r[i,:,:,:], axis=0)
TOT_180_r = np.zeros((TOT_180.shape[0],3,227,227),dtype=np.uint8)
for i in range(TOT_180.shape[0]):
temp = TOT_180[i,0,:,:]
temp2 = cv2.resize(temp,(227,227))
TOT_180_r[i,2,:,:] = temp2-m[0]
temp = TOT_180[i,1,:,:]
temp2 = cv2.resize(temp,(227,227))
TOT_180_r[i,1,:,:] = temp2-m[1]
temp = TOT_180[i,2,:,:]
temp2 = cv2.resize(temp,(227,227))
TOT_180_r[i,0,:,:] = temp2-m[2]
TOT_180_r[i,:,:,:] = np.expand_dims(TOT_180_r[i,:,:,:], axis=0)
TOT_270_r = np.zeros((TOT_270.shape[0],3,227,227),dtype=np.uint8)
for i in range(TOT_270.shape[0]):
temp = TOT_270[i,0,:,:]
temp2 = cv2.resize(temp,(227,227))
TOT_270_r[i,2,:,:] = temp2-m[0]
temp = TOT_270[i,1,:,:]
temp2 = cv2.resize(temp,(227,227))
TOT_270_r[i,1,:,:] = temp2-m[1]
temp = TOT_270[i,2,:,:]
temp2 = cv2.resize(temp,(227,227))
TOT_270_r[i,0,:,:] = temp2-m[2]
TOT_270_r[i,:,:,:] = np.expand_dims(TOT_270_r[i,:,:,:], axis=0)
np.savez('Data_7_97j_seg', TOT_0_r=TOT_0_r, TOT_90_r=TOT_90_r, TOT_180_r=TOT_180_r, TOT_270_r=TOT_270_r, GL=GL, y1=y1, Tray=Tray, BVZ=BVZ)
print(y1)
lab1=np.sum(y1==1)
lab2=np.sum(y1==2)
lab3=np.sum(y1==3)
lab4=np.sum(y1==4)
print(np.sum(y1==1))
print(np.sum(y1==2))
print(np.sum(y1==3))
print(np.sum(y1==4))
# print(GL)
print(GL.shape[0])
print(TOT_0_r.shape[0])
#########################################################
#### CNN feature extraction
#########################################################
#CNN_model = convnet('alexnet',weights_path="alexnet_weights.h5", heatmap=False)
CNN_model = convnet('alexnet24',weights_path="./tmp/weights_alex_7_1.hdf5", heatmap=False)
# CNN_model = convnet('vgg_16',weights_path="vgg16_weights.h5", heatmap=False)
from keras import backend as K
get_dense_features = K.function([CNN_model.layers[0].input, K.learning_phase()],[CNN_model.layers[31].output]) #33
print ("Feature extraction...")
X_0 = np.zeros((97, INPUT_LEN, INPUT_DIM))
X_90 = np.zeros((97, INPUT_LEN, INPUT_DIM))
X_180 = np.zeros((97, INPUT_LEN, INPUT_DIM))
X_270 = np.zeros((97, INPUT_LEN, INPUT_DIM))
y = np.zeros((97, INPUT_LEN, OUTPUT_LEN))
for newcounter in range(0, 97):
for inp_len in range (0, 22):
current_frame = np.zeros((1,3,227,227))
current_frame = np.array(np.expand_dims(TOT_0_r[newcounter*22+inp_len,:,:,:], axis=0))
X_0[newcounter, inp_len] = np.array(get_dense_features([current_frame,0]))[0]
current_frame = np.zeros((1,3,227,227))
current_frame = np.array(np.expand_dims(TOT_90_r[newcounter*22+inp_len,:,:,:], axis=0))
X_90[newcounter, inp_len] = np.array(get_dense_features([current_frame,0]))[0]
current_frame = np.zeros((1,3,227,227))
current_frame = np.array(np.expand_dims(TOT_180_r[newcounter*22+inp_len,:,:,:], axis=0))
X_180[newcounter, inp_len] = np.array(get_dense_features([current_frame,0]))[0]
current_frame = np.zeros((1,3,227,227))
current_frame = np.array(np.expand_dims(TOT_270_r[newcounter*22+inp_len,:,:,:], axis=0))
X_270[newcounter, inp_len] = np.array(get_dense_features([current_frame,0]))[0]
y[newcounter, inp_len, GL[newcounter*22+10]-1] = 1
print (newcounter)
#########################################################
#### Preparing the data for training the LSTM model
#########################################################
######################################################
#### LSTM Model
#########################################################
INPUT_LEN = 22
INPUT_DIM = 4096
OUTPUT_LEN = 4
ind_tot = range(97)
random.seed(16)
rand00=random.sample(ind_tot,97)
#print('@@@@@@@@@@@@@@@@@@@@@@@@')
#print(rand00)
for j in range(j):
TD_model = Sequential()
#TD_model.add(SimpleRNN(256,return_sequences=True, input_dim=INPUT_DIM, input_length=INPUT_LEN)) #256
TD_model.add(LSTM(256,return_sequences=True, input_dim=INPUT_DIM, input_length=INPUT_LEN)) #256
TD_model.add(Dropout(0.5))
TD_model.add(LSTM(256,return_sequences=True ))
TD_model.add(Dropout(0.5))
TD_model.add(TimeDistributed(Dense(OUTPUT_LEN)))
# TD_model.add(TimeDistributed(Dense(2)))
# TD_model.add(Dense(4))
TD_model.add(Activation('softmax'))
# sgd = SGD(lr=0.01, decay=0.005 , momentum=0.9, nesterov=True)
# TD_model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])
BATCH_SIZE = 32
ind_test_0 = []
ind_train_0 = []
ind_val_0 = []
if j!= 4 :
ind_test_0 = rand00[j*20:(j+1)*20]
ind_val_0 = rand00[82:97]
else :
ind_test_0 = range(0,20)
ind_val_0 = range(82,97)
ind_train_0 = list(set(ind_tot) - set(ind_test_0) - set(ind_val_0))
ind_test_90 = []
ind_train_90 = []
ind_val_90 = []
ind_test_180 = []
ind_train_180 = []
ind_val_180 = []
ind_test_270 = []
ind_train_270 = []
ind_val_270 = []
ind_test_90 = ind_test_0
ind_val_90 = ind_val_0
ind_train_90 = ind_train_0
ind_test_180 = ind_test_0
ind_val_180 = ind_val_0
ind_train_180 = ind_train_0
ind_test_270 = ind_test_0
ind_val_270 = ind_val_0
ind_train_270 = ind_train_0
newYtrain = []
newYtest = []
newYval = []
X_train1 = []
X_val1 = []
X_test1 = []
X_test1 = np.concatenate((X_0[ind_test_0,:,:],X_90[ind_test_90,:,:],X_180[ind_test_180,:,:],X_270[ind_test_270,:,:]),axis=0)
X_val1 = np.concatenate((X_0[ind_val_0,:,:],X_90[ind_val_90,:,:],X_180[ind_val_180,:,:],X_270[ind_val_270,:,:]),axis=0)
X_train1 = np.concatenate((X_0[ind_train_0,:,:],X_90[ind_train_90,:,:],X_180[ind_train_180,:,:],X_270[ind_train_270,:,:]),axis=0)
newYtest=np.concatenate((y[ind_test_0],y[ind_test_90],y[ind_test_180],y[ind_test_270]),axis=0)
newYval=np.concatenate((y[ind_val_0],y[ind_val_90],y[ind_val_180],y[ind_val_270]),axis=0)
newYtrain=np.concatenate((y[ind_train_0],y[ind_train_90],y[ind_train_180],y[ind_train_270]),axis=0)
ny= np.zeros((newYval.shape[0], INPUT_LEN))
for nf in range(0, newYval.shape[0]):
for kf in range(22) :
for df in range(4) :
if newYval[nf,kf,df]==1:
ny[nf,kf]=df
print(ny)
checkpointer = ModelCheckpoint(filepath="./tmp/weights_SNP_7_eachframe_print.hdf5", verbose=1, save_best_only=True)
sgd = SGD(lr=0.01, decay=0.005, momentum=0.9, nesterov=True)
TD_model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])
TD_model.fit(X_train1, newYtrain, batch_size=32, nb_epoch=50, validation_data = (X_val1, newYval), callbacks=[checkpointer], sample_weight= None)
# LSTM_model.save_weights("./tmp/final_weights_SNP1.hdf5",overwrite=True)
TD_model.save_weights("./tmp/final_weights_SNP_7_eachframe_print.hdf5",overwrite=True)
loss_and_metrics = TD_model.evaluate(X_val1, newYval)
classes= np.array([])
classes = TD_model.predict_classes(X_val1)
score = TD_model.evaluate(X_val1, newYval, batch_size=BATCH_SIZE)
print(score)
result1= score[1]
classes = TD_model.predict_classes(X_test1)
score = TD_model.evaluate(X_test1, newYtest, batch_size=BATCH_SIZE)
print(score)
result2= score[1]
# print(newYtest)
nytt=np.zeros(newYtest.shape[0])
for nf in range(0, newYtest.shape[0]):
for df in range(4):
if newYtest[nf,0,df]==1:
nytt[nf]=df
ny= np.zeros((newYtest.shape[0], INPUT_LEN))
# ny=np.zeros(newYtest.shape[0])
for nf in range(0, newYtest.shape[0]):
for kf in range(22) :
for df in range(4) :
if newYtest[nf,kf,df]==1:
ny[nf,kf]=df
TD_model.load_weights("./tmp/weights_SNP_7_eachframe_print.hdf5")
TD_model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])
loss_and_metrics = TD_model.evaluate(X_test1, newYtest)
classes = TD_model.predict_classes(X_test1)
clas=np.ones(newYtest.shape[0])*6
clas1=np.ones(22)*6
for nf in range(0, newYtest.shape[0]):
count=0
for kf in range(22) :
clas1[kf]=classes[nf,kf]
clas1=np.array(clas1, dtype=np.int64)
counts = np.bincount(clas1)
clas[nf]=np.argmax(counts)
print('$$$$$$$$$$$$$$$$$$$$$')
print(clas)
print('$$$$$$$$$$$$$$$$$$$$$')
print(nytt)
score = TD_model.evaluate(X_test1, newYtest, batch_size=BATCH_SIZE)
print(score)
# result3= score[1]
f_matrix3 = confusion_matrix(nytt, clas)
print(np.where(clas != nytt)[0])
print((np.where(clas != nytt)[0])%20)
print((np.where(clas != nytt)[0])/20)
ind_test_0=np.array(ind_test_0)
print(ind_test_0[(np.where(classes != ny)[0])%20])
print(Tray[ind_test_0[(np.where(classes != ny)[0])%20]])
print(BVZ[ind_test_0[(np.where(classes != ny)[0])%20]])
print(y1[ind_test_0[(np.where(classes != ny)[0])%20]]-1)
#print(ny)
#print(classes)
loss_and_metrics = TD_model.evaluate(X_val1, newYval)
classes = TD_model.predict_classes(X_val1)
print(classes)
score = TD_model.evaluate(X_val1, newYval, batch_size=BATCH_SIZE)
print(score)
print(f_matrix3)
cl_p= TD_model.predict_proba(X_test1)
print(cl_p[10])
print(cl_p[10,1])
with open('prob_print.csv','a') as f:
writer = csv.writer(f)
writer.writerow(('BVZ','Tray_label_single','Plant_label_single'))
for nff in range (0,20):
writer.writerow((BVZ[rand00[nff]],Tray[rand00[nff]],y1[rand00[nff]]))
for nt in range (0,22):
writer.writerow((cl_p[nff,nt]))