The model provides a plausible pose along with uncertainty estimates for the joints.
Microsoft presented Flow-based 3D Avatar Generation from Sparse Observations (FLAG) – a model that creates a 3D representation of a human body based on the head-mounted display's (HMD) signals.
HMDs' signal streams are usually limited to the head and hands poses, so Microsoft is working to generate realistic and faithful avatar poses of the whole body.
The method also provides "a probabilistic mapping from observations to the latent space from which we can generate a plausible pose along with uncertainty estimates for the joints."
Check out the research here and don't forget to join our new Reddit page, our new Telegram channel, follow us on Instagram and Twitter, where we are sharing breakdowns, the latest news, awesome artworks, and more.