Something about autoencoder-like network in Tensorflow, however the net structure is a crazy scribble. 😂 Just practice using file queue feature in TF.

autoencoder.py

import tensorflow as tf


def read_image(filename_queue):
    reader = tf.WholeFileReader()
    key, value = reader.read(filename_queue)

    img = tf.image.decode_jpeg(value)
    img.set_shape((256, 512, 3))

    img = tf.image.convert_image_dtype(img, dtype=tf.float32)

    img = img * 2 - 1
    img_a = img[:, :256, :]
    img_b = img[:, 256:, :]

    return img_a, img_b


def input_pipeline(batch_size, num_epochs=None):
    filenames = tf.train.match_filenames_once('./datasets/facades/train/*.jpg')

    filename_queue = tf.train.string_input_producer(
        filenames, num_epochs=num_epochs, shuffle=True)

    example_a, example_b = read_image(filename_queue)

    min_after_dequeue = 100
    capacity = min_after_dequeue + 3 * batch_size

    example_batch = tf.train.shuffle_batch(
        [example_a, example_b], batch_size=batch_size, capacity=capacity,
        min_after_dequeue=min_after_dequeue)

    return example_batch

''''''

img_a, img_b = input_pipeline(32, num_epochs=10)

x = tf.layers.conv2d(img_a, 16, [3, 3], padding='same', activation=tf.nn.relu)
x = tf.layers.max_pooling2d(x, pool_size=[2, 2], strides=2)
x = tf.layers.conv2d(x, 8, [3, 3], padding='same', activation=tf.nn.relu)
x = tf.layers.max_pooling2d(x, pool_size=[2, 2], strides=2)
x = tf.layers.conv2d(x, 8, [3, 3], padding='same', activation=tf.nn.relu)
x = tf.layers.max_pooling2d(x, pool_size=[2, 2], strides=2)

x = tf.layers.conv2d(x, 8, [3, 3], padding='same', activation=tf.nn.relu)
x = tf.layers.conv2d_transpose(x, 8, [3, 3], [2, 2], padding='same')
x = tf.layers.conv2d(x, 8, [3, 3], padding='same', activation=tf.nn.relu)
x = tf.layers.conv2d_transpose(x, 8, [3, 3], [2, 2], padding='same')
x = tf.layers.conv2d(x, 16, [3, 3], padding='same', activation=tf.nn.relu)
x = tf.layers.conv2d_transpose(x, 16, [3, 3], [2, 2], padding='same')
out = tf.layers.conv2d(x, 3, [3, 3], padding='same', activation=tf.nn.sigmoid)

cost = tf.reduce_mean(tf.pow(out - img_b, 2))
optimizer = tf.train.RMSPropOptimizer(0.0002).minimize(cost)

''''''

sess = tf.Session()
sess.run([tf.global_variables_initializer(), tf.local_variables_initializer()])

coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)

try:
    while not coord.should_stop():
        _, c = sess.run([optimizer, cost])
        print(c)
except tf.errors.OutOfRangeError:
    print('Done training -- epoch limit reached')
finally:
    coord.request_stop()

coord.join(threads)
sess.close()