-
Notifications
You must be signed in to change notification settings - Fork 0
/
Resnet_processing.py
166 lines (109 loc) · 6.01 KB
/
Resnet_processing.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
from __future__ import absolute_import, division, print_function, unicode_literals
import tensorflow as tf
from tensorflow.keras.applications.resnet50 import ResNet50
from tensorflow.keras.models import Model, Sequential
from batch_maker import StimMaker, all_test_shapes
from tensorflow.keras.layers import Dense, Input, BatchNormalization, Flatten
from tensorflow.keras.optimizers import Adam
import numpy as np
from tensorflow.keras.utils import to_categorical
########################################################################################################################
# Decoder Layer Generator
########################################################################################################################
def Minimodel(n_hidden, input_shape):
mod = Sequential()
mod.add(Flatten(input_shape=input_shape[1:]))
mod.add(Dense(n_hidden, activation='elu'))
mod.add(Dense(2, activation='softmax'))
return mod
########################################################################################################################
# Training
########################################################################################################################
def train_loop(base_model, sample_layers, input_maker, train_n_batches, batch_size, NAME, ID_MODEL, lr=1e-6):
print('\rTraining...')
print('\rModel ID:'+NAME+str(ID_MODEL))
path = './logdir' + NAME + str(ID_MODEL)
train_summary_writer = tf.summary.create_file_writer(path)
minimodels = [Minimodel(n_hidden, base_model.layers[L].output_shape) for L in sample_layers]
loss_object = tf.keras.losses.SparseCategoricalCrossentropy()
metrics = tf.keras.metrics.SparseCategoricalAccuracy()
with train_summary_writer.as_default():
optimizer = Adam(lr=lr)
for iteration in range(train_n_batches):
print('Batch n°'+str(iteration) + ' / ' + str(train_n_batches))
#Generate train data:
batch_data, batch_labels = input_maker.generate_Batch(batch_size, [0, 0, 1, 0], noiseLevel=.1,
normalize=False,
fixed_position=None)
RES_outputs = base_model(batch_data)
for i, mini in enumerate(minimodels):
with tf.GradientTape() as tape:
logits = mini(RES_outputs[i])
loss = loss_object(batch_labels, logits)
gradients = tape.gradient(loss, mini.trainable_variables)
optimizer.apply_gradients(zip(gradients, mini.trainable_variables))
tf.summary.scalar('___loss'+str(i), loss, step=optimizer.iterations)
acc = metrics(batch_labels, logits)
metrics.reset_states()
if iteration % 25 == 0:
print('Decoder n°'+str(i)+' training accuracy:'+str(acc.numpy()*100)+'%')
tf.summary.scalar('___acc' + str(i), acc, step=optimizer.iterations)
# Saving the minimodels
for i, mini in enumerate(minimodels):
for layer in mini.layers:
layer.trainable = False
mini.save(path + '/Minimodel_L='+str(sample_layers[i])+'.h5')
########################################################################################################################
# Testing
########################################################################################################################
def test_loop(base_model, sample_layers, input_maker, SHAPES, test_set_size, NAME, ID_MODEL):
print('\rTesting...')
path= './logdir' + NAME + str(ID_MODEL)
#Recovering saved models
minimodels = [tf.keras.models.load_model(path+'/Minimodel_L='+str(L)+'.h5') for L in sample_layers]
N_tests = len(SHAPES)
results = np.zeros((N_tests, len(sample_layers)))
metrics = tf.keras.metrics.SparseCategoricalAccuracy()
for s,shapematrix in enumerate(SHAPES):
print('\rshapematrix ={}'.format(str(shapematrix)))
batch_data, batch_labels = input_maker.generate_Batch(test_set_size, [0, 0, 0, 1], noiseLevel=.1,
normalize=False,
fixed_position=None, shapeMatrix=shapematrix)
RES_outputs = base_model(batch_data)
for i, mini in enumerate(minimodels):
logits = mini(RES_outputs[i])
metrics.reset_states()
results[s,i] = metrics(batch_labels, logits)
print('Finished testing for this model, numpy saving\n'.format())
np.save(path + '/results', results)
np.save(path + '/shape_list', SHAPES)
return results
########################################################################################################################
#Main:
########################################################################################################################
IMG_SIZE = 224 # 224
IMG_SHAPE = (IMG_SIZE, IMG_SIZE, 3)
n_models = 10
NAME = 'ResNet_test'
#Training parameters:
train_n_batches = 10000
batch_size = 64
test_set_size = 64*100
n_hidden = 512
### BASE MODEL DEFINITION (UNIQUE):
# activation layers before next conv : 4, 38, 80, 142, 174
sample_layers = [4, 38, 80, 142, 174]
model = ResNet50(include_top=False, weights='imagenet', input_shape=IMG_SHAPE)
outputs = [model.get_layer(model.layers[L].name).output for L in sample_layers]
intermediate_layer_model = Model(inputs=model.input, outputs=outputs)
# input definition
input_maker = StimMaker(imSize=(IMG_SIZE, IMG_SIZE), shapeSize=19, barWidth=2)
# test shapes definition
SHAPES = all_test_shapes()
### MAIN LOOP
for m in range(n_models):
### Model is defined and saved via train_loop functions
## Uniquely identified by the NAME and MODEL_ID
ID_MODEL = m
train_loop(intermediate_layer_model, sample_layers, input_maker, train_n_batches, batch_size, NAME, ID_MODEL)
test_loop(intermediate_layer_model, sample_layers, input_maker, SHAPES, test_set_size, NAME, ID_MODEL)