search

moonbow721 / DPoser

Official implementation of the paper "DPoser: Diffusion Model as Robust 3D Human Pose Prior"
MIT License
46 stars 1 forks source link

Question about smpl-x motion denoising #1

Closed GrigorySamokhin closed 5 months ago

GrigorySamokhin commented 5 months ago

Hello! Thanks for your amazing work. Is it possible to apply motion denoising to SMPL-X model (with 71 joints) instead of 21 in your demo? And how good it will work if im going to apply it to in-the-wild animations?

moonbow721 commented 5 months ago

Hi there!

Thank you so much for your interest in our work, it's great to see your enthusiasm! Regarding your question about the SMPL-X model's 71 joints, let me clarify a bit. The SMPL-X model actually encompasses 54 local poses, including 21 for the body, 30 for the hands, and 3 for the face. In our DPoser demo, we focused on the local body prior, which accounts for the 21 joints you've noticed.

Now, for your intriguing idea about applying DPoser to full-body tasks like motion denoising, it's definitely a path worth exploring! To do this, we'd need to train a model specifically for the whole body, leveraging datasets like SMPL-X in AMASS. It's an exciting prospect, and while we haven't specifically tested motion denoising in wild environments, our experiments with human mesh recovery in such settings have yielded promising results.

I encourage you to experiment with your own data for motion denoising. You can easily do this by adjusting the file_path parameter in our demo. Feel free to reach out if you have more questions or exciting updates!

GrigorySamokhin commented 5 months ago

I apologize for the late response. Thanks for the detailed answer! I will definitely try