FLAG: Flow-based 3D Avatar Generation from Sparse Observations

CVPR 2022 |

To represent people in mixed reality applications for collaboration and communication, we need to generate realistic and faithful avatar poses. However, the signal streams that can be applied for this task from head-mounted devices (HMDs) are typically limited to head pose and hand pose estimates. While these signals are valuable, they are an incomplete representation of the human body, making it challenging to generate a faithful full-body avatar. We address this challenge by developing a flow-based generative model of the 3D human body from sparse observations, wherein we learn not only a conditional distribution of 3D human pose, but also a probabilistic mapping from observations to the latent space from which we can generate a plausible pose along with uncertainty estimates for the joints. We show that our approach is not only a strong predictive model, but can also act as an efficient pose prior in different optimization settings where a good initial latent code plays a major role.

 

Flow-based 3D Avatar Generation from Sparse Observations

Microsoft Swiss Joint Research Center – Day 2 | Computer vision and Mixed Reality The Applied Machine Learning Days channel features talks and performances from the Applied Machine Learning Days. AMLD is one of the largest machine learning & AI events in Europe, focused specifically on the applications of machine learning and AI, making it particularly interesting to industry and academia.