-
Notifications
You must be signed in to change notification settings - Fork 0
/
dense_net.py
58 lines (47 loc) · 2 KB
/
dense_net.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
import train_model
import tensorflow as tf
import numpy as np
import keras.api._v2.keras as K
class ResidualLayer(K.layers.Layer):
def __init__(self, ffnn, **kwargs):
super().__init__(**kwargs)
self.ffnn = K.models.Sequential(ffnn)
def call(self, inputs, *args, **kwargs):
return tf.concat((self.ffnn(inputs), inputs), axis=-1)
def get_dense_net():
inputs = K.layers.Input(shape=(128, 128, 1))
x = K.layers.Conv2D(filters=256, kernel_size=1, padding="SAME", activation="linear")(inputs)
x = K.layers.BatchNormalization()(x)
x = K.layers.Activation(tf.nn.leaky_relu)(x)
concatenated_inputs = x
for f in [128] * 7:
x = K.layers.Conv2D(filters=64, kernel_size=1, padding="SAME", activation="linear")(x)
x = K.layers.BatchNormalization()(x)
x = K.layers.Activation(tf.nn.leaky_relu)(x)
x = K.layers.Conv2D(filters=f, kernel_size=4, padding="SAME", activation="linear")(x)
x = K.layers.BatchNormalization()(x)
x = K.layers.Activation(tf.nn.leaky_relu)(x)
concatenated_inputs = K.layers.Concatenate()([concatenated_inputs, x])
x = K.layers.Conv2D(filters=64, kernel_size=1, padding="SAME", activation="linear")(x)
x = K.layers.BatchNormalization()(x)
x = K.layers.Activation(tf.nn.leaky_relu)(x)
x = K.layers.Conv2D(3, 3, activation="sigmoid", padding="SAME")(x)
dense_net = K.models.Model(inputs=[inputs], outputs=[x])
return dense_net
if __name__ == "__main__":
print("0%")
with open('cache/X_train.npy', 'rb') as f:
X_train = np.load(f)
print("33%")
with open('cache/X_val.npy', 'rb') as f:
X_val = np.load(f)
print("66%")
with open('cache/X_test.npy', 'rb') as f:
X_test = np.load(f)
print("100%")
dense_net = get_dense_net()
dense_net.compile(
tf.optimizers.legacy.Adam(1e-3),
loss=tf.losses.MeanSquaredError()
)
train_model.train_dnn(X_train, X_val, dense_net, "dense_net", batch_size=16, save_only_weights=True)