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

Central location for transforms #904

Closed
matsumotosan opened this issue Oct 11, 2022 · 7 comments · Fixed by #905
Closed

Central location for transforms #904

matsumotosan opened this issue Oct 11, 2022 · 7 comments · Fixed by #905
Labels
help wanted Extra attention is needed won't fix This will not be worked on

Comments

@matsumotosan
Copy link
Contributor

Data transform classes for various models are included in their own folders in pl_bolts/models. It may be better to have them in pl_bolts/transforms instead.

Issue discussed with @otaj

@matsumotosan matsumotosan added the help wanted Extra attention is needed label Oct 11, 2022
@senarvi
Copy link
Contributor

senarvi commented Oct 31, 2022

I was going to suggest the same thing. Also, at least SimCLR and MoCo can use the same transforms. Maybe we want to provide some transforms that can be used to replicate the paper results on some dataset, but do we want to provide a separate set of transforms for every model and every dataset?

Related to this, I feel like it would be good to somehow standardize the format in which the data is processed (data types, dimensions of the tensors), for different tasks, so that any classification model can be used with any classification dataset and transforms, and same for detection and SSL models. I don't know how much this is true already.

@matsumotosan
Copy link
Contributor Author

In general, I think the existing transforms are based on what the original papers specified (for whatever datasets they used). If a model by default uses transforms from another (i.e. BYOL using SimCLR transforms), I think we can just import SimCLRTransforms for BYOL instead of creating a copy and calling it BYOLTransforms. I think creating transforms for every model for every dataset might be too much though.

I'm all for standardizing how transforms are set up. I'm almost done moving everything over to pl_bolts/transforms so I'll let you know when I'm finished with that.

@otaj
Copy link
Contributor

otaj commented Nov 2, 2022

Hi, @matsumotosan, that would be great, thank you for kicking it off!

@matsumotosan
Copy link
Contributor Author

All self-supervised transforms have been moved #905

@stale
Copy link

stale bot commented Jan 7, 2023

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

@stale stale bot added the won't fix This will not be worked on label Jan 7, 2023
@matsumotosan
Copy link
Contributor Author

Commenting so this issue doesn't get closed.

All changes have been made in #905. Waiting for CI to be fixed so we can finish testing.

@stale stale bot removed the won't fix This will not be worked on label Jan 8, 2023
@matsumotosan matsumotosan linked a pull request Jan 9, 2023 that will close this issue
8 tasks
@stale
Copy link

stale bot commented Mar 18, 2023

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

@stale stale bot added the won't fix This will not be worked on label Mar 18, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
help wanted Extra attention is needed won't fix This will not be worked on
Projects
None yet
Development

Successfully merging a pull request may close this issue.

3 participants