Digital makeup using deep learning method

• Main Author: Aung, Thandar
• Other Authors: He Ying
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering::Theory of computation
• Online Access: https://hdl.handle.net/10356/157704
• Institution: Nanyang Technological University

In the modern age, the world is like a village day by day as the technology is being developed as a skyrocket. As the world is being developed, people might need to continuously develop their own country to be the best growth by their own technologies. Hence, people are getting tired of working life and stress. Thus, the entertainment section plays as one of the important roles among other sections to maintain working-life balanced. As an entertainment, sharing portraits and self-portraits are one of the trending criteria among each other. Since time is precious for everyone, it is such a waste to beautify the pictures manually. It is time consuming task if users do manual tuning for the photos, even for the experienced users. Makeup look is also distinct in style according to worldwide facial look (for e.g., Asian face shape, Western face shape and so on). Thus, it includes a variety of local styles or cosmetics, such as eye shadow, lipstick, foundation, etc. Obtaining and transferring such different face shape to present style, and sensitive cosmetics information is infeasible ways of transport. The problem will be dealt by combining the two. In a dual-domain system, there is a global domain-level loss and a local instance-level loss. Beauty GAN is an input/output Generative Adversarial Network. Discriminators ensure domain-level transfer that distinguishes output photos from real samples getting from domains. This project will discuss how the existing various deep learning models can be performed in transferring makeup style while maintaining the face’s uniqueness. An overview of various neural network algorithms such as BeautyGAN, Convolutional neural network (CNN) that has been implemented in existing paper will be included. Finally, experimental results will be reviewed in terms of various existing deep learning models.