3D Digital Avatars with Machine Learning

Creating 3D digital garments is an active area of research due to a large number of applications in many fields, including fashion design, e-commerce, virtual try-on, and video games. The traditional approaches to this problem use physics-based simulation techniques to model how clothing deforms, but the high computational cost required at run time hinders the deployment of these techniques to real-world applications. Alternatively, recent methods based on Machine Learning are able to reconstruct 3D garments directly from images and to infer how 3D garments deform when worn by arbitrary body shapes. This has opened the door to the democratization of digital clothing, with a direct impact on video games, to improve to the visual fidelity of 3D characters; online shopping, to enable to virtually try on clothes in online stores; and fashion, to speed up the design process to create. In this talk, I will introduce recent state-of-the-art techniques for digital avatars introduced by our lab, including friendly descriptions of the fundamental parts of this exciting line of research in Computer Graphics and Machine Learning. Plenty of cool video results and nice 3D renders!