How to generate the specified 3D motion

roykapon / MAS

The official implementation of the paper "MAS: Multiview Ancestral Sampling for 3D Motion Generation Using 2D Diffusion"

MIT License

97 stars 3 forks source link

How to generate the specified 3D motion #7

Closed XingliangJin closed 3 months ago

XingliangJin commented 5 months ago

In your paper, the visual comparison experiment seems to require the generation of specified 3D motion. However, the method uses 2d diffusion with unconditional training. So, how do you generate the specified 3D motion?

roykapon commented 3 months ago

Hello @XingliangJin :) I'm not sure I understood your question properly. The visual comparison experiment is unconditioned, meaning we just took some random generations (not necessarily of the same type) and compared them.

XingliangJin commented 3 months ago

Thank you！I mistakenly thought that the comparison methods showed the same type of motions.