Hierarchical Graph-Convolutional Variational Autoencoding for Generative Modelling of Human Motion
Abstract: Models of human motion focus either on trajectory prediction or action classification but rarely both. The marked heterogeneity and intricate compositionality of human motion render each task vulnerable to the data degradation and distributional shift common to real-world scenarios. A sufficiently expressive generative model of action could in theory enable data conditioning and distributional resilience within a unified framework applicable to both tasks as well as facilitate data synthesis. We propose a novel architecture for generating a holistic model of action based on hierarchical variational autoencoders and deep graph convolutional neural networks. We show this Hierarchical Graph-convolutional Variational AutoEncoder (HG-VAE) to be capable of detecting out-of-distribution data, and imputing missing data by gradient ascent on the model's posterior, facilitating better downstream discriminative learning. We show that scaling to greater stochastic depth generates better likelihoods independently of model capacity. We further show that the efficient hierarchical dependencies HG-VAE learns enable the generation of coherent conditioned actions and robust definition of class domains at the top level of abstraction. We trained and evaluated on H3.6M and the largest collection of open source human motion data, AMASS.
Submission Length: Regular submission (no more than 12 pages of main content)
Assigned Action Editor: ~Masha_Itkina1
Submission Number: 2194
Loading