CV-ASDF2Net/model_ASDF2Net.py at main · mqalkhatib/CV-ASDF2Net · 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
'''
Shallow to deep network
'''

import tensorflow as tf
import cvnn.layers as complex_layers
from tensorflow import keras
from SAR_utils import *
import tensorflow as tf
from tensorflow.keras.layers import Conv2D, Flatten, Dense, Dropout, Input

###############################################################################
from SAR_utils import cmplx_SE_Block


def ASDF2Net(X_cmplx, num_classes):

    cmplx_inputs = complex_layers.complex_input(shape=(X_cmplx.shape[1:]))

    # Shallow Path
    c0 = complex_layers.ComplexConv3D(16, activation='cart_relu', kernel_size=(3,3,3), padding="same")(cmplx_inputs)

    # Mid Path
    c1 = complex_layers.ComplexConv3D(16, activation='cart_relu', kernel_size=(3,3,3), padding="same")(cmplx_inputs)
    c1 = complex_layers.ComplexConv3D(16, activation='cart_relu', kernel_size=(3,3,3), padding="same")(c1)


    # Deep Path
    c2 = complex_layers.ComplexConv3D(16, activation='cart_relu', kernel_size=(3,3,3), padding="same")(cmplx_inputs)
    c2 = complex_layers.ComplexConv3D(16, activation='cart_relu', kernel_size=(3,3,3), padding="same")(c2)
    c2 = complex_layers.ComplexConv3D(16, activation='cart_relu', kernel_size=(3,3,3), padding="same")(c2)

    # Attenstion Block
    features_concat = tf.concat([c0, c1, c2], axis = 4)
    se = cmplx_SE_Block_3D(features_concat, se_ratio = 8)
    se = cmplx_SE_Block_3D(se, se_ratio = 8)
    se = cmplx_SE_Block_3D(se, se_ratio = 8)

    # Flatenning Features
    features_concat_flat = complex_layers.ComplexFlatten()(se)


    # Dense and Dropout
    c3 = complex_layers.ComplexDense(128, activation='cart_relu')(features_concat_flat)
    c3 = complex_layers.ComplexDropout(0.25)(c3)
    c4 = complex_layers.ComplexDense(64, activation='cart_relu')(c3)
    c4 = complex_layers.ComplexDropout(0.25)(c4)

    # Prediction
    predict = complex_layers.ComplexDense(num_classes,activation="softmax_real_with_abs")(c4)


    model = tf.keras.Model(inputs=[cmplx_inputs], outputs=predict)
    model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['accuracy'])

    return model

#######################################################################################################
def cmplx_SE_Block_3D(xin, se_ratio = 8):
    # Squeeze Path
    xin = tf.transpose(xin, perm=[0, 1, 2, 4, 3])
    xin_gap =  GlobalCmplxAveragePooling3D(xin)
    sqz = complex_layers.ComplexDense(xin.shape[-1]//se_ratio, activation='cart_relu')(xin_gap)

    # Excitation Path
    excite1 = complex_layers.ComplexDense(xin.shape[-1], activation='cart_sigmoid')(sqz)

    out = tf.keras.layers.multiply([xin, excite1])
    out = tf.transpose(out, perm=[0, 1, 2, 4, 3])

    return out


def GlobalCmplxAveragePooling3D(inputs):
    inputs_r = tf.math.real(inputs)
    inputs_i = tf.math.imag(inputs)

    output_r = tf.keras.layers.GlobalAveragePooling3D()(inputs_r)
    output_i = tf.keras.layers.GlobalAveragePooling3D()(inputs_i)

    if inputs.dtype == 'complex' or inputs.dtype == 'complex64' or inputs.dtype == 'complex128':
           output = tf.complex(output_r, output_i)
    else:
           output = output_r

    return output