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

Add CLI for finetuning Segment Anything #836

Open
anwai98 opened this issue Jan 17, 2025 · 2 comments
Open

Add CLI for finetuning Segment Anything #836

anwai98 opened this issue Jan 17, 2025 · 2 comments
Assignees
Labels
discussion Do we want to add this feature adnd how?

Comments

@anwai98
Copy link
Contributor

anwai98 commented Jan 17, 2025

I think it would be nice to have a simple CLI for finetuning SAM / micro-sam models (and have heard of some requests from other users as well)! What do you think @constantinpape?

PS. I will only come back to this early next month!

@anwai98 anwai98 added the discussion Do we want to add this feature adnd how? label Jan 17, 2025
@constantinpape
Copy link
Contributor

Yes, I agree that this would be good to have.

@anwai98 anwai98 self-assigned this Jan 17, 2025
@anwai98
Copy link
Contributor Author

anwai98 commented Jan 20, 2025

Another feature request: (I've also been thinking about this for a while now)

  • It would make sense to not store model checkpoints in the current working directory, and allow users to specify their directory (using save_root). From what I understand, this should be updated! (https://github.com/computational-cell-analytics/micro-sam/blob/master/micro_sam/sam_annotator/training_ui.py#L244-L250)
  • Review the advanced parameters:
    • Initial model: we can fetch the value from Configuration and set it to default, or display a template. I've seen users get confused about it!
    • Checkpoint: Is there a reason to put a default value there? I would call it something like Checkpoint path and provide a template for a filepath.
    • Can we support an export script which stores the SAM model state and decoder_state in one filepath? (this would simplify our design to export all relevant components for the finetuned model)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
discussion Do we want to add this feature adnd how?
Projects
None yet
Development

No branches or pull requests

2 participants