Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Question for embedding dimension #16

Closed
JosephKKim opened this issue May 12, 2022 · 2 comments
Closed

Question for embedding dimension #16

JosephKKim opened this issue May 12, 2022 · 2 comments

Comments

@JosephKKim
Copy link

Thanks for the wonderful work @Mathux!
I really enjoyed reading paper and also testing it on my environment,
while looking at your code and paper, I found out that you've set dimensionality of embedding for VAE to 256 and I'm wondering if there's any reason for this.

Thanks,
Joseph

@Mathux
Copy link
Owner

Mathux commented May 12, 2022

Hello @JosephKKim,

Yes, the dimensionality of the latent space is fixed at 256. This exact number is a bit arbitrary, but the intuition may be that it corresponds to a "motion embedding". So, in my opinion, it should be larger than a simple "pose embedding". Some papers use 64 or 128 for pose embedding, so I just took the next higher power of two.

Feel free to experiment with another value if you wish :)

@JosephKKim
Copy link
Author

JosephKKim commented May 12, 2022

Thank you so much for the fast reply... I didn't expect that I would get a reply this fast.. :)
Okay, I see your point that "motion" would need more embedding dimension than that of the pose.
I was trying to improve ACTOR by using properties from VAEs.. so that was my motivation for this issue.

Once again, thank you so much!

P.S. I've also enjoyed your paper TEMOS (Congrats!), since I've read MotionCLIP paper and wondered how you and your group will make use of other modalities.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants