Project Analysis - Color Transfer / GAN

Project Analysis - Color Transfer

Git ref - Link

• XYZ to CIE-LAB color space conversion
• skimage.color.rgb2lab(rgb[, illuminant, …])
• Conversion from the sRGB color space (IEC 61966-2-1:1999) to the CIE Lab colorspace under the given illuminant and observer.

	# -*- coding: utf-8 -*-
	"""
	Automatically generated by Colaboratory.
	"""
	#Upload files to colab
	from google.colab import files
	files.upload()
	#https://github.com/emilwallner/Coloring-greyscale-images/blob/master/Alpha-version/alpha_version_notebook.ipynb
	from keras.layers import Conv2D, UpSampling2D, InputLayer, Conv2DTranspose
	from keras.layers import Activation, Dense, Dropout, Flatten
	from tensorflow.keras.layers import BatchNormalization
	from keras.models import Sequential
	from keras.preprocessing.image import ImageDataGenerator, array_to_img, img_to_array, load_img
	from skimage.color import rgb2lab, lab2rgb, rgb2gray, xyz2lab
	from skimage.io import imsave
	import numpy as np
	import os
	import random
	import tensorflow as tf
	# Get images
	import cv2
	image_old = img_to_array(load_img('2.jpg'))
	image_old = np.array(image_old, dtype=float)
	image = cv2.resize(image_old, (400, 400),
	interpolation = cv2.INTER_NEAREST)
	X = rgb2lab(1.0/255*image)[:,:,0]
	Y = rgb2lab(1.0/255*image)[:,:,1:]
	from IPython.display import Image
	Y /= 128
	X = X.reshape(1, 400, 400, 1)
	Y = Y.reshape(1, 400, 400, 2)
	from google.colab.patches import cv2_imshow
	Image(image)
	# Building the neural network
	model = Sequential()
	model.add(InputLayer(input_shape=(None, None, 1)))
	model.add(Conv2D(8, (3, 3), activation='relu', padding='same', strides=2))
	model.add(Conv2D(8, (3, 3), activation='relu', padding='same'))
	model.add(Conv2D(16, (3, 3), activation='relu', padding='same'))
	model.add(Conv2D(16, (3, 3), activation='relu', padding='same', strides=2))
	model.add(Conv2D(32, (3, 3), activation='relu', padding='same'))
	model.add(Conv2D(32, (3, 3), activation='relu', padding='same', strides=2))
	model.add(UpSampling2D((2, 2)))
	model.add(Conv2D(32, (3, 3), activation='relu', padding='same'))
	model.add(UpSampling2D((2, 2)))
	model.add(Conv2D(16, (3, 3), activation='relu', padding='same'))
	model.add(UpSampling2D((2, 2)))
	model.add(Conv2D(2, (3, 3), activation='tanh', padding='same'))
	# Finish model
	model.compile(optimizer='rmsprop',loss='mse')
	model.fit(x=X,
	y=Y,
	batch_size=1,
	epochs=1000)
	print(model.evaluate(X, Y, batch_size=1))
	output = model.predict(X)
	output *= 128
	# Output colorizations
	cur = np.zeros((400, 400, 3))
	cur[:,:,0] = X[0][:,:,0]
	cur[:,:,1:] = output[0]
	imsave("img_result.png", lab2rgb(cur))
	imsave("img_gray_version.png", rgb2gray(lab2rgb(cur)))
	from IPython.display import Image
	Image('img_gray_version.png')
	Image('img_result.png')

view raw alphaversion.py hosted with ❤ by GitHub

Project #2 - deep koalarization

Paper  - link

Key Notes

• High-level feature extraction using a pre-trained model (Inception-ResNetv2) to enhance the coloring process.

• Fusion The fusion layer takes the feature vector from Inception, replicates it HW/8*8 times and attaches it to the feature volume outputted by the encoder along the depth axis

More Reads

Color Spaces

Paper - Colorization Using ConvNet and GAN

• Colorization is a popular image-to-image translation problem
• We implemented two models for the task: ConvNet and conditional-GAN and found that GAN can generate better results both quantitatively and qualitatively
• ConvNet and GAN. Both models are design to take either grayscale or edge-only images and produce color (RGB) images.

Types of GAN

Attgan

More Reads

Controlling Colors of GAN-Generated and Real Images via Color Histograms

Keep Exploring!!!