Sketch-to-Color Image with GANs

Unsupervised Learning is a trending research field of artificial intelligence, which aims to interpret and understand the hidden structure of the data. However, the development of deep learning with Generative Adversarial Networks (GANs) creates more possibilities for unsupervised learning. GAN is a category of Neural Networks, which are mostly applied to generating images. In this paper, how GAN was implemented to help with sketch-to-color translation is illustrated. In order to achieve this goal, data-preprocessing is implemented first. Then, the model is trained for 65 epochs, and the performance of the model is improved by virtue of loss functions and optimizers. In the end, a proper User Interface (GUI) is designed to have a full application, and people could turn any sketch picture they want into a colored image.