CBAM,代填坑

​x
def CBAM(input, reduction):
    """
    @Convolutional Block Attention Module
    """

    _, width, height, channel = input.get_shape()  # (B, W, H, C)

    # channel attention
    x_mean = tf.reduce_mean(input, axis=(1, 2), keepdims=True)   # (B, 1, 1, C)
    x_mean = tf.layers.conv2d(x_mean, channel // reduction, 1, activation=tf.nn.relu, name='CA1')  # (B, 1, 1, C // r)
    x_mean = tf.layers.conv2d(x_mean, channel, 1, name='CA2')   # (B, 1, 1, C)

    x_max = tf.reduce_max(input, axis=(1, 2), keepdims=True)  # (B, 1, 1, C)
    x_max = tf.layers.conv2d(x_max, channel // reduction, 1, activation=tf.nn.relu, name='CA1', reuse=True)
    # (B, 1, 1, C // r)
    x_max = tf.layers.conv2d(x_max, channel, 1, name='CA2', reuse=True)  # (B, 1, 1, C)

    x = tf.add(x_mean, x_max)   # (B, 1, 1, C)
    x = tf.nn.sigmoid(x)        # (B, 1, 1, C)
    x = tf.multiply(input, x)   # (B, W, H, C)

    # spatial attention
    y_mean = tf.reduce_mean(x, axis=3, keepdims=True)  # (B, W, H, 1)
    y_max = tf.reduce_max(x, axis=3, keepdims=True)  # (B, W, H, 1)
    y = tf.concat([y_mean, y_max], axis=-1)     # (B, W, H, 2)
    y = tf.layers.conv2d(y, 1, 7, padding='same', activation=tf.nn.sigmoid)    # (B, W, H, 1)
    y = tf.multiply(x, y)  # (B, W, H, C)

    return y