Tensorflow Implementation of Bilinear Additive Upsampling

tf_bilinear_additive_upsampling.py

import tensorflow as tf

"""
Author : @MikiBear_
Tensorflow Implementation of Bilinear Additive Upsampling.
Reference : https://arxiv.org/abs/1707.05847
"""
def bilinear_additive_upsampling(x, to_channel_num, name):
    from_channel_num = x.get_shape().as_list()[3]
    assert from_channel_num % to_channel_num == 0
    channel_split = from_channel_num / to_channel_num

    new_shape = x.get_shape().as_list()
    new_shape[1] *= 2
    new_shape[2] *= 2
    new_shape[3] = to_channel_num
    
    upsampled_x = tf.image.resize_images(x, new_shape[1:3])

    output_list = []
    for i in range(to_channel_num):
        splited_upsampled_x = upsampled_x[:,:,:,i*channel_split:(i+1)*channel_split]
        output_list.append(tf.reduce_sum(splited_upsampled_x, axis = -1))

    output = tf.stack(output_list, axis = -1)

    return output

if __name__ == '__main__':
    x = tf.ones([20, 100, 100, 20])
    y = bilinear_additive_upsampling(x, 5, '0')
    with tf.Session() as sess:
        init_op = tf.global_variables_initializer()
        sess.run(init_op)
        new_x = sess.run(y)
        print(new_x)

In python3 it should be
channel_split = from_channel_num // to_channel_num