Computer Vision News - April 2023

7 SinMDM: Single Motion Diffusion learned motion motifs to synthesize the motion of the legs or vice versa. Moving forward, the team anticipates that other researchers could build upon their work by leveraging the key concepts that made the diffusion model effective in learning from a single animation and exploring new applications beyond those already introduced. Apart from contributing to this project, Inbal has been actively pursuing her master’s degree in computer science and works at Microsoft. “ It was a big honor for me to work with such smart people on this paper, ” she reveals. “ I’ve learned a lot from them, including how they work, how to approach things, and how to get the best paper done in time. It’s been great fun! ” give you that. The network can also generate long motions. Let’s say we train on a two-second dance sequence, we can then generate a full minute of the same dance, and it would look continuous rather than repetitive. Another application is style transfer, where we train the model on a style motion, like someone walking happily, and then transfer that style to a newmotion using a harmonization technique. ” Other applications include temporal composition , where given the prefix or center of a motion, the network can synthesize the rest of it, introducing diverse outputs, and spatial composition, where the network uses certain body parts as input and generates others. For example, the motion of the arms is based on a reference motion, and the model uses