convolution_ae.py

# -*- coding: utf-8 -*-
"""convolution_AE.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1CjmgX06DElqcb78skfaDOkJoiv6NHFuW
"""

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
from tensorflow.contrib.layers import fully_connected
import cv2

mnist = input_data.read_data_sets("/MNIST_data/", one_hot=True)

x = tf.placeholder(tf.float32, shape = [None, 784])
X = tf.reshape(x, [-1, 28, 28, 1])

initializer = tf.contrib.layers.variance_scaling_initializer()
#kernels
ew1 = tf.Variable(initializer([3,3,1,32]), dtype = tf.float32)
ew2 = tf.Variable(initializer([3,3,32,64]), dtype = tf.float32)
ew3 = tf.Variable(initializer([3,3,64,128]), dtype = tf.float32)

dw1 = tf.Variable(initializer([3,3,128,64]), dtype = tf.float32)
dw2 = tf.Variable(initializer([3,3,64,32]), dtype = tf.float32)
dw3 = tf.Variable(initializer([3,3,32,16]), dtype = tf.float32)

dw = tf.Variable(initializer([3,3,16,1]), dtype = tf.float32)

#biases
eb1 = tf.Variable(tf.zeros(32))
eb2 = tf.Variable(tf.zeros(64))
eb3 = tf.Variable(tf.zeros(128))

db1 = tf.Variable(tf.zeros(64))
db2 = tf.Variable(tf.zeros(32))
db3 = tf.Variable(tf.zeros(16))

db = tf.Variable(tf.zeros(1))

strides = 1 #convolution
k = 2 #pooling

#model

conv1 = tf.nn.conv2d(X, ew1, strides = [1, strides, strides, 1], padding = 'SAME')
conv1 = tf.nn.bias_add(conv1, eb1)
conv1 = tf.nn.leaky_relu(conv1)
conv1 = tf.nn.max_pool(conv1, ksize = [1, k, k, 1], strides = [1, k, k, 1], padding= 'SAME')

conv2 = tf.nn.conv2d(conv1, ew2, strides = [1, strides, strides, 1], padding = 'SAME')
conv2 = tf.nn.bias_add(conv2, eb2)
conv2 = tf.nn.leaky_relu(conv2)
conv2 = tf.nn.max_pool(conv2, ksize = [1, k, k, 1], strides = [1, k, k, 1], padding= 'SAME')

conv3 = tf.nn.conv2d(conv2, ew3, strides = [1, strides, strides, 1], padding = 'SAME')
conv3 = tf.nn.bias_add(conv3, eb3)
conv3 = tf.nn.leaky_relu(conv3)
encoder = tf.nn.max_pool(conv3, ksize = [1, k, k, 1], strides = [1, k, k, 1], padding= 'SAME')
print(encoder.get_shape())

upsample1 = tf.image.resize_images(encoder, size = (7,7), method = tf.image.ResizeMethod.NEAREST_NEIGHBOR)
conv4 = tf.nn.conv2d(upsample1, dw1, strides = [1, strides, strides, 1], padding = 'SAME')
conv4 = tf.nn.bias_add(conv4, db1)
conv4 = tf.nn.leaky_relu(conv4)

upsample2 = tf.image.resize_images(conv4, size = (14,14), method = tf.image.ResizeMethod.NEAREST_NEIGHBOR)
conv5 = tf.nn.conv2d(upsample2, dw2, strides = [1, strides, strides, 1], padding = 'SAME')
conv5 = tf.nn.bias_add(conv5, db2)
conv5 = tf.nn.leaky_relu(conv5)

upsample3 = tf.image.resize_images(conv5, size = (28,28), method = tf.image.ResizeMethod.NEAREST_NEIGHBOR)
conv6 = tf.nn.conv2d(upsample3, dw3, strides = [1, strides, strides, 1], padding = 'SAME')
conv6 = tf.nn.bias_add(conv6, db3)
conv6 = tf.nn.leaky_relu(conv6)

decoder = tf.nn.conv2d(conv6, dw, strides = [1, strides, strides, 1], padding = 'SAME')
decoder = tf.nn.bias_add(decoder, db)
decoder = tf.nn.sigmoid(decoder)
oned = tf.reshape(decoder,[-1, 784])
print(decoder.get_shape())

with tf.device('/gpu:0'): 
    lr = 0.0001
    #loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(labels = x, logits = oned))
    loss = tf.reduce_mean(tf.square(decoder - X))
    optimizer = tf.train.AdamOptimizer(lr)
    train = optimizer.minimize(loss)
    init = tf.global_variables_initializer()

num_epoch = 20
batch_size = 150
num_test_images = 10

with tf.Session() as sess:
    sess.run(init)
    for epoch in range(num_epoch):
        num_batches = mnist.train.num_examples // batch_size
        for i in range(num_batches):
            X_batch,_ = mnist.train.next_batch(batch_size)
            noise_factor = 0.5
            sess.run(train, feed_dict={x:X_batch})
        train_loss = loss.eval(feed_dict = {x:X_batch})
        print("epoch {} loss {}".format(epoch+1, train_loss))
    results = decoder.eval(feed_dict={x:mnist.test.images[:num_test_images]})
    f,a=plt.subplots(2,10,figsize=(20,4))
    for i in range(num_test_images):
        a[0][i].imshow(np.reshape(mnist.test.images[i],(28,28)))
        a[1][i].imshow(np.reshape(results[i],(28,28)))