id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
cord-027119-zazr8uj5	Taif, Khasrouf	Cast Shadow Generation Using Generative Adversarial Networks	2020-05-25		.txt	text/plain	4029	221	56	Generative Adversarial Networks have been implemented widely to perform graphical tasks, as it requires minimum to no human interaction, which gives GANs a great advantage over conventional deep learning methods, such as image-to-image translation with single D, G semi-supervised model [7] or unsupervised dual learning [26] . We apply image-to-image translation to our own image set to generate correct cast shadows for 3D rendered images in a semi-supervised manner using colour labels. We start with the assumption that GANs can generate both soft and hard shadows on demand, using colour labels and given a relatively small training image set. This paper explored a framework based on conditional GANs using a pix2pix Tensorflow port to perform computer graphic functions, by instructing the network to successfully generate shadows for 3D rendered images given training images paired with conditional colour labels.	./cache/cord-027119-zazr8uj5.txt	./txt/cord-027119-zazr8uj5.txt