-
Notifications
You must be signed in to change notification settings - Fork 231
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
feat: GaussianMixtureConditional #239
Conversation
Co-authored-by: Jinming Liu <[email protected]>
sorry for the late response, I was busy with other things. Can you give me push access to your fork for a quick fix before merge |
Thanks, though the "allow edits by maintainers" option seems already on my side. I've also sent an invitation to my fork if this doesn't work. |
Correct, I messed-up and created a new branch, thanks |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
partially tested, approving. TODO (fracape): train, test and upload model
|
||
return likelihood | ||
|
||
def forward( |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Have you tried with cheng2020-anchor+gmm or cheng2020-attn+gmm (and compared results)? I think you assembled the model outside compressai for testing. Can you add the model? I'd like to try a training in same conditions as the other models (vimeo90k, etc.) and also test the regression with K=1.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sorry I must have missed this comment somehow and just came across it after 4 months.
I developed this GMM variant for some other model architecture so haven't tried on the exactly original Cheng2020.
I do used to work with the lab that published the Cheng2020 paper. Let me go ask a few guys if they have some pretrained models.
#238, Gaussian Mixture Model implementation which is used in Cheng2020 paper.
Implementation details follow the official code but I changed CDF generation into a parallel algorithm so that it's easier to work with parallel context models.
Also modified
rans_interface
to allow a torch tensor to be CDF, or ~80% of execution time is wasted on type conversion.