Learning Motion Manifolds with Convolutional Autoencoders
Created on Sept. 9, 2015, 11:47 a.m.
This year I got both my papers accepted! At SIGGRAPH Asia I presented Learning Motion Manifolds with Convolutional Autoencoders. This paper is about learning a manifold over the space of human motion using deep convolutional autoencoders. This is a useful task for many aspects of Animation Research and Machine Learning because it represents a strong prior belief about human motion data. You can download the paper here.
Abstract: We present a technique for learning a manifold of human motion data using Convolutional Autoencoders. Our approach is capable of learning a manifold on the complete CMU database of human motion. This manifold can be treated as a prior probability distribution over human motion data, which has many applications in animation research, including projecting invalid or corrupt motion onto the manifold for removing error, computing similarity between motion using geodesic distance along the manifold, and interpolation of motion along the manifold to avoid blending artefacts.