ResNet/ResNet.py at main · TensorFlow-ML-Architectures/ResNet · 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
from keras import layers
from tensorflow import keras

class ResNet:
    def __int__(self):
        self.dims = None
        self.num_classes = None
        self.model = None
        self.inputs = None

    def model_create(self, dims, num_classes):
        self.dims = dims
        self.num_classes = num_classes
        self.inputs = keras.Input(shape=self.dims)

        # Image augmentation block
        rot = 0.1
        flip = "horizontal"
        data_augmentation = keras.Sequential(
            [
                layers.RandomFlip(flip),
                layers.RandomRotation(rot),
            ]
        )
        x = data_augmentation(self.inputs)

        # re-scale
        x = layers.Rescaling(1.0 / 255)(x)

        # Entry block
        x = layers.Conv2D(32, 3, strides=2, padding="same")(x)
        x = layers.BatchNormalization()(x)
        x = layers.Activation("relu")(x)

        x = layers.Conv2D(64, 3, padding="same")(x)
        x = layers.BatchNormalization()(x)
        x = layers.Activation("relu")(x)

        previous_block_activation = x  # Set aside residual

        for size in [128, 256, 512, 728]:
            x = layers.Activation("relu")(x)
            x = layers.SeparableConv2D(size, 3, padding="same")(x)
            x = layers.BatchNormalization()(x)

            x = layers.Activation("relu")(x)
            x = layers.SeparableConv2D(size, 3, padding="same")(x)
            x = layers.BatchNormalization()(x)

            x = layers.MaxPooling2D(3, strides=2, padding="same")(x)

            # Project residual
            residual = layers.Conv2D(size, 1, strides=2, padding="same")(
                previous_block_activation
            )
            x = layers.add([x, residual])  # Add back residual
            previous_block_activation = x  # Set aside next residual

        x = layers.SeparableConv2D(1024, 3, padding="same")(x)
        x = layers.BatchNormalization()(x)
        x = layers.Activation("relu")(x)

        x = layers.GlobalAveragePooling2D()(x)
        if self.num_classes == 2:
            activation = "sigmoid"
            units = 1
        else:
            activation = "softmax"
            units = self.num_classes

        x = layers.Dropout(0.5)(x)
        outputs = layers.Dense(units, activation=activation)(x)

        self.model = keras.Model(self.inputs, outputs)

        self.model.summary()