main.py

""" 
DISCLAIMER:
This code has been written in 2017 in the optic
of my 'quick-n-dirty' Deep Learning series
on Medium (@juliendespois) to show the 
concept. Please do not judge me by the 
quality of the code. 
¯\_(ツ)_/¯
"""

from keras.datasets import mnist
import numpy as np
import matplotlib.pyplot as plt
from keras.layers import Input, Dense, Convolution2D, MaxPooling2D, UpSampling2D
from keras.models import Model
from keras.callbacks import TensorBoard
from keras.models import model_from_json
import cv2
import sys
import os
from random import shuffle, randint, choice

# Generates random image with squares and circles
def get_random_image():
	img_size = 100
	size = 25
	nb_shapes = 5

	xy = lambda: randint(0,100)

	# Create a white image
	img = np.zeros((img_size,img_size,3), np.uint8)
	cv2.rectangle(img,(0,0),(img_size,img_size),(122,122,122) ,-1)

	grey_img = np.copy(img)

	# Adds some shapes
	for i in range(nb_shapes):
		x0, y0 = xy(), xy()
		is_rect = choice((True,False))
		if is_rect:
			cv2.rectangle(img,(x0,y0),(x0+size,y0+size),(255,0,0) ,-1)
			cv2.rectangle(grey_img,(x0,y0),(x0+size,y0+size),(255,255,255) ,-1)
		else:
			cv2.circle(img,(x0,y0), size/2, (0,0,255), -1)
			cv2.circle(grey_img,(x0,y0), size/2, (255,255,255), -1)

	return cv2.resize(img,(48,48)), cv2.resize(grey_img,(48,48))

# Creates the dataset
def get_dataset(display=False):
	# Show what the dataset looks like
	if display:
		colorImg, grey_img = get_random_image()
		img = np.hstack((colorImg, grey_img))
		cv2.imshow("Dataset",cv2.resize(img,(200,100)))
		cv2.waitKey(0)
		cv2.destroyAllWindows()

	#for i in range

	x_train, x_test, y_train, y_test = [], [], [], []

	# Add training examples
	for i in range(10000):
		colorImg, grey_img = get_random_image()
		grey_img = cv2.cvtColor(grey_img, cv2.COLOR_RGB2GRAY)
		x_train.append(grey_img.astype('float32')/255.)
		y_train.append(colorImg.astype('float32')/255.)

	# Add test examples
	for i in range(1000):
		colorImg, grey_img = get_random_image()
		grey_img = cv2.cvtColor(grey_img, cv2.COLOR_RGB2GRAY)
		x_test.append(grey_img.astype('float32')/255.)
		y_test.append(colorImg.astype('float32')/255.)

	# Reshape
	x_train = np.array(x_train).reshape((-1,48,48,1))
	x_test = np.array(x_test).reshape((-1,48,48,1))
	y_train = np.array(y_train).reshape((-1,48,48,3))
	y_test = np.array(y_test).reshape((-1,48,48,3))

	return x_train, y_train, x_test, y_test

# Creates the Convolutional Auto Encoder
def get_model():
	input_img = Input(shape=(48, 48, 1))
	x = Convolution2D(16, 3, 3, activation='relu', border_mode='same', dim_ordering='tf')(input_img)
	x = MaxPooling2D((2, 2), border_mode='same', dim_ordering='tf')(x)
	x = Convolution2D(32, 3, 3, activation='relu', border_mode='same', dim_ordering='tf')(input_img)
	x = MaxPooling2D((2, 2), border_mode='same', dim_ordering='tf')(x)
	x = Convolution2D(64, 3, 3, activation='relu', border_mode='same', dim_ordering='tf')(x)
	encoded = MaxPooling2D((2, 2), border_mode='same', dim_ordering='tf')(x)
	#6x6x32 -- bottleneck
	x = UpSampling2D((2, 2), dim_ordering='tf')(encoded)
	x = Convolution2D(32, 3, 3, activation='relu', border_mode='same', dim_ordering='tf')(x)
	x = UpSampling2D((2, 2), dim_ordering='tf')(x)
	x = Convolution2D(16, 3, 3, activation='relu', border_mode='same', dim_ordering='tf')(x)
	decoded = Convolution2D(3, 3, 3, activation='relu', border_mode='same', dim_ordering='tf')(x)

	#Create model
	autoencoder = Model(input_img, decoded)
	return autoencoder

# Trains the model for 10 epochs
def train_model():
	# Load dataset
	print("Loading dataset...")
	x_train_gray, x_train, x_test_gray, x_test = get_dataset()

	# Create model description
	print("Creating model...")
	model = get_model()
	model.compile(optimizer='rmsprop', loss='binary_crossentropy',metrics=['accuracy'])

	# Train model
	print("Training model...")
	model.fit(x_train_gray, x_train, nb_epoch=10, batch_size=148, shuffle=True, validation_data=(x_test_gray, x_test), callbacks=[TensorBoard(log_dir='/tmp/tb', histogram_freq=0, write_graph=False)])

	# Evaluate loaded model on test data
	print("Evaluating model...")
	score = model.evaluate(x_train_gray, x_train, verbose=0)
	print "%s: %.2f%%" % (model.metrics_names[1], score[1]*100)

	# Serialize model to JSON
	print("Saving model...")
	model_json = model.to_json()
	with open("model.json", "w") as json_file:
	    json_file.write(model_json)
	    
	# Serialize weights to HDF5
	print("Saving weights...")
	model.save_weights("model.h5")

# Tests the model and shows results
def test_model():
	# Load JSON model description
	with open('model.json', 'r') as json_file:
		modelJSON = json_file.read()
	
	# Build model from JSON description
	print("Loading model...")
	model = model_from_json(modelJSON)

	# Load weights
	print("Loading weights...")
	model.load_weights("model.h5")

	_, _, x_test_gray, x_test = get_dataset()
	x_test_gray = x_test_gray[:10]
	x_test = x_test[:10]

	print("Making predictions...")
	predictions = model.predict(x_test_gray)
	x_test_gray = [cv2.cvtColor(img,cv2.COLOR_GRAY2RGB) for img in x_test_gray]

	img = np.vstack((np.hstack(x_test_gray), np.hstack(predictions), np.hstack(x_test)))

	cv2.imshow("Input - Reconstructed - Ground truth",cv2.resize(img, (img.shape[1], img.shape[0])))
	cv2.waitKey(0)
	cv2.destroyAllWindows()


if __name__ == "__main__":
	arg = sys.argv[1] if len(sys.argv) == 2 else None 
	if arg is None:
		print "Need argument"
	elif arg == "train":
		train_model()
	elif arg == "test":
		test_model()
	elif arg == "dataset":
		get_dataset(True)
	else:
		print "Wrong argument"