def smart_res_block()

in utils.py [0:0]

• 62 lines of code
• 5 McCabe index (conditional complexity)

def smart_res_block(
        inp,
        weights,
        reuse,
        scope,
        downsample=True,
        adaptive=True,
        stop_batch=False,
        upsample=False,
        label=None,
        act=tf.nn.leaky_relu,
        dropout=False,
        train=False,
        **kwargs):
    gn1 = weights[scope + '_res_c1']
    gn2 = weights[scope + '_res_c2']
    c1 = smart_conv_block(
        inp,
        weights,
        reuse,
        scope + '_res_c1',
        use_stride=False,
        activation=None,
        extra_bias=True,
        label=label,
        **kwargs)

    if dropout:
        c1 = tf.layers.dropout(c1, rate=0.5, training=train)

    c1 = act(c1)
    c2 = smart_conv_block(
        c1,
        weights,
        reuse,
        scope + '_res_c2',
        use_stride=False,
        activation=None,
        use_scale=True,
        extra_bias=True,
        label=label,
        **kwargs)

    if adaptive:
        c_bypass = smart_conv_block(
            inp,
            weights,
            reuse,
            scope +
            '_res_adaptive',
            use_stride=False,
            activation=None,
            **kwargs)
    else:
        c_bypass = inp

    res = c2 + c_bypass

    if upsample:
        res_shape = tf.shape(res)
        res_shape_list = res.get_shape()
        res = tf.image.resize_nearest_neighbor(
            res, [2 * res_shape_list[1], 2 * res_shape_list[2]])
    elif downsample:
        res = tf.nn.avg_pool(res, (1, 2, 2, 1), (1, 2, 2, 1), 'VALID')

    res = act(res)

    return res