Function conv_block

src/models/blocks.py:12–78

basic convolution block Args: inputs: input to the conv layer conv_type: type of convolution. "conv" for Conv2D and "ds" for depthwise separable convolution filters: number of filters kernel_size: kernel size used in convolution strides: strides used

(inputs,
               conv_type="conv",
               filters=64,
               kernel_size=(3, 3),
               strides=(1, 1),
               padding="same",
               relu=True,
               upsampling=False,
               up_sample_size=2,
               skip_layer=None
               )

import sys

def conv_block(inputs,
               conv_type="conv",
               filters=64,
               kernel_size=(3, 3),
               strides=(1, 1),
               padding="same",
               relu=True,
               upsampling=False,
               up_sample_size=2,
               skip_layer=None
               ):
    """
    basic convolution block
    Args:
        inputs: input to the conv layer
        conv_type: type of convolution. "conv" for Conv2D and "ds" for depthwise separable convolution
        filters: number of filters
        kernel_size: kernel size used in convolution
        strides: strides used in convolution
        padding: padding used in convolution
        relu: if relu is active or not. 
        upsampling: if need to upsample
        skip_layer: if skip layer is added
    Returns:
        out: output of basic convolution block
    """
    if conv_type == "ds":
        out = tf.keras.layers.SeparableConv2D(filters=filters,
                                              kernel_size=kernel_size,
                                              padding=padding,
                                              strides=strides)(inputs)
    elif conv_type == "conv":
        out = tf.keras.layers.Conv2D(filters=filters,
                                     kernel_size=kernel_size,
                                     padding=padding,
                                     strides=strides)(inputs)
    else:
        sys.exit("Wrong choice of convolution type.")
    
    out = tf.keras.layers.BatchNormalization()(out)
    
    if relu:
        out = tf.keras.activations.relu(out)
    
    if upsampling:
        out = tf.keras.layers.UpSampling2D(size=(up_sample_size, up_sample_size),
                                           data_format="channels_last")(out)
        skip_layer = tf.keras.layers.Conv2D(filters=out.shape[3], kernel_size=(1, 1), padding='same', strides=(1, 1))(skip_layer)
        out = tf.keras.layers.concatenate([out, skip_layer], axis=3)
        if conv_type == "ds":
            out = tf.keras.layers.SeparableConv2D(filters=filters,
                                                kernel_size=kernel_size,
                                                padding=padding,
                                                strides=strides)(out)
        elif conv_type == "conv":
            out = tf.keras.layers.Conv2D(filters=filters,
                                        kernel_size=kernel_size,
                                        padding=padding,