import numpy as np

import tensorflow as tf

def gaussian(x, n_c=3):

gaussian_5x5 = tf.constant(np.array([

[1, 4, 6, 4, 1],

[4, 16, 25, 16, 4],

[6, 24, 36, 24, 6],

[4, 16, 25, 16, 4],

[1, 4, 6, 4, 1],

], dtype=np.float32) / 256.0)

gaussian_weights = tf.reshape(gaussian_5x5, (5, 5, 1, 1))

gaussian5x5_rgb = tf.concat([

tf.nn.conv2d(tf.expand_dims(x[:, :, :, 0], axis=3), gaussian_weights, strides=[1, 2, 2, 1], padding='VALID'),

tf.nn.conv2d(tf.expand_dims(x[:, :, :, 1], axis=3), gaussian_weights, strides=[1, 2, 2, 1], padding='VALID'),

tf.nn.conv2d(tf.expand_dims(x[:, :, :, 2], axis=3), gaussian_weights, strides=[1, 2, 2, 1], padding='VALID')

], axis=3)

return gaussian5x5_rgb

def laplacian(x, n_c=3):

laplacian_operator = tf.constant(np.array([

[0, -1, 0],

[-1, 4, -1],

[0, -1, 0],

], dtype=np.float32) / 4.0)

lap_weights = tf.reshape(laplacian_operator, (3, 3, 1, 1))

laplacian3x3_rgb = tf.concat([

tf.nn.conv2d(tf.expand_dims(x[:, :, :, 0], axis=3), lap_weights, strides=[1, 2, 2, 1], padding='VALID'),

tf.nn.conv2d(tf.expand_dims(x[:, :, :, 1], axis=3), lap_weights, strides=[1, 2, 2, 1], padding='VALID'),

tf.nn.conv2d(tf.expand_dims(x[:, :, :, 2], axis=3), lap_weights, strides=[1, 2, 2, 1], padding='VALID')

], axis=3)

return laplacian3x3_rgb

def bilinear(x, n_c=3, factor=0.75):

new_size = tf.to_int32(tf.cast(tf.shape(x)[1:3], tf.float32) * factor)

return tf.image.resize_bilinear(x, new_size)