-
Notifications
You must be signed in to change notification settings - Fork 76
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
[No positional encoding for position and direction] #17
Comments
Hi @RaymondJiangkw ,
I guess "officially acknowledged" is a little bit too strong as a claim, since the SIREN based net is only used for 2D image regression in the paper. For novel view synthesis with NeRF, they still use ReLU MLPs with positional encoding. |
Hi, @DeanChan
Of course, it's strange, haha, and that's the reason why there's a double quote. In the paper of original NeRF team, they acknowledged the effectiveness of SIREN when dealing with the problem of Coordinate-Based Neural Representations, even though they only applied it to 2D image regression, and positional encoding is exactly targeted at solving the same issue. I'm sorry if you feel my words are not accurate, but what I want to emphasize is that -GAN choosing SIREN as an alternative way is plausible and has its theoretical foundation. Sorry for my informal claim. 😳 |
Many thanks for your quick reply 😄 It's much clearer now. I just found a repo applying SIREN to NeRF--, which may support the effectiveness of SIREN. In their implementation, there's no positional encoding for SIREN-based NeRF either. See frameworks.py#L28-L38 |
Dear marco,
Thanks for open source the code! I am curious about why positional encoding is not involved in pi-GAN's implementation.
As in
https://github.com/marcoamonteiro/pi-GAN/blob/master/generators/generators.py#L49
, could you please be so kind to tell me the reason of removing positional encoding totransformed_points
(xyz) andray_directions
(d)?Thank you for your time!
The text was updated successfully, but these errors were encountered: