Skip to content

Port explicit denoising methods from fmri-fm.archive#7

Open
har5h1l wants to merge 4 commits intoMedARC-AI:mainfrom
har5h1l:dev/denoising
Open

Port explicit denoising methods from fmri-fm.archive#7
har5h1l wants to merge 4 commits intoMedARC-AI:mainfrom
har5h1l:dev/denoising

Conversation

@har5h1l
Copy link
Copy Markdown

@har5h1l har5h1l commented Oct 28, 2025

This PR ports the work on the explicit denoising methods in models_mae.py and its demo notebook from fmri-fm.archive.

denoising_demo.ipynb currently doesn't work, as it's still in the process of being updated for the newer repo codebase.

This is not intended to be pushed to main, there should be a new branch called dev/denoising made when merging.

- Add denoising_demo.ipynb with updated imports and model loading
- Fix patch_embed references in models_mae.py to use encoder.patchify
- Support both new pretrained models and fallback to scratch initialization
- Update notebook to work with new flat_mae module structure
…the denoising_demo notebook to be compatible.
- Implement dynamic module reloading in denoising_demo.ipynb for updated model imports.
- Add pixel_mask creation for visualization in denoising_demo.ipynb.
- Update print statements to include detailed output shapes and processing times.
- Modify models_mae.py to ensure compatibility with input data types and improve patch handling.
- Refactor patch_mask creation to accommodate both 2D and 3D cases, ensuring dtype consistency.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant