You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello, great work! But I have a question about the FSQ employed in your work.
I noticed that each group has 10 dimensions, the size of codebook is 625 (5^4) described in your paper, why not 5^10?
The text was updated successfully, but these errors were encountered:
Hello, in fact, this is also a hyperparameter. The reason for not using higher dimensions is: 1. We want to keep the dimensions as small as possible to ensure that the discrete bitrate does not exceed the continuous one, and 2. The performance improvement brought by trying larger dimensions is actually very limited. Therefore, we used the parameters from the paper, and more careful searching and tuning may lead to further improvements.
Hello, great work! But I have a question about the FSQ employed in your work.
I noticed that each group has 10 dimensions, the size of codebook is 625 (5^4) described in your paper, why not 5^10?
The text was updated successfully, but these errors were encountered: