Backtracking tests and moving modules around for better organization#36
Conversation
…which internally serializes and deserializes the manifest, so for dataset to and from config the only source of truth would be the file_state attribute.
…implified serialization logic
…ual_stain_flow into dev-add-patch-dataset
…atial dimension behavior. Fix inappropriate property decorator on callable functions.
…, fix imports and tests
MattsonCam
left a comment
MattsonCam
left a comment
There was a problem hiding this comment.
Adding these tests was a great idea—nice work! I don't see anything in the implementation that looks likely to introduce regressions, and the tests look reasonable.
I did notice that some of the import paths have changed due to the code being moved in this pr. I think this may break 1.modular_unet_example in the examples folder. It might also be worth creating versioned releases (or tags) for this repository so users can pin a specific version when training virtual staining models, especially as the API evolves.
…taset - Replaced PatchDataset with CropImageDataset and BaseImageDataset for improved data handling. - Updated data loading logic to accommodate new dataset structure. - Adjusted visualization code to reflect changes in dataset and patch generation. - Renumbered notebook cells for clarity and consistency.
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
Great call here, i have added a compatibility shim to prevent breaking API changes in this PR. |
…style implementation
…example data acquisition. Now all example training and augmentation uses the newly acquired CPJUMP1 A549 images. Removing old data acquisition script
|
Updated outdated training and augmentation example notebooks to use the new CPJUMP1 dataset. |
|
Thanks @MattsonCam for reviewing, merging now! |
2 participants
This PR primes a series of future refactors whose collective goal would be improving readability and expand-ability of models, and ultimately laying foundation for future feature addition of allowing conditioning by a separate variable while computing convolutions on images/feature maps (conditioning is required for diffusion modelling).
This PR does the following:
blockssub package without changing behavior, adjust imports