Multi-modal learning of human pose representation for conditional motion synthesis

The goal of the project is to generate realistic human movement in 3D animations. This is important to make movement animations in games and movies appear real. Typically, creating high quality animations is a resource and time consuming process that requires the participation of human actors in motion capture sessions. In this work, we present a data-driven approach that aims to generate novel animations based on a library of past motion capture recordings that can make generating high quality animations low cost and fast by eliminating the need to record human actors. Our proposed method leverages state-of-the-art machine learning algorithms and Unity’s high-quality motion capture data to generate 3D human movement.

Faculty Supervisor:

Mark Coates

Student:

Antonios Valkanas

Partner:

Unity Technologies

Discipline:

Engineering - computer / electrical

Sector:

Professional, scientific and technical services

University:

McGill University