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

DLMBL 2024 notebook #114

Merged
merged 15 commits into from
Aug 8, 2024
Merged

DLMBL 2024 notebook #114

merged 15 commits into from
Aug 8, 2024

Conversation

edyoshikun
Copy link
Contributor

@edyoshikun edyoshikun commented Jul 25, 2024

This PR adds the exercise for the DL@MBL 2024 course (Part 1) for the image translation.
It also reorganizes the demo folders for clearer organization for the #119 wiki.

Major changes to our img2img translation example:

  • Added the comparison to the pretrained VSCyto2D model
  • Predictions and cellpose segmentations are computed and saved into OME-Zarr for further comparison (part 2 exercise)
  • Compute pixel-based and segmentation-based metrics and stores them to a dataframe for further comparison.

@ziw-liu ziw-liu added the documentation Improvements or additions to documentation label Jul 29, 2024
- adding evaluation of pretrained vs course trained model
- saving of the predictions
- saving of the pixl based metrics and segmentation metrics
- evaluation metrics pixel and segmentation
- saving predictions for further evaluation
@edyoshikun edyoshikun marked this pull request as ready for review August 2, 2024 15:50
@edyoshikun
Copy link
Contributor Author

this PR should be merged before #119.

@edyoshikun edyoshikun requested a review from ziw-liu August 2, 2024 15:52
@edyoshikun
Copy link
Contributor Author

edyoshikun commented Aug 2, 2024

I am also updating cellpose to 3.0.10 in this PR. We were already using 3.0.1 so this shouldn't cause issues. @ziw-liu

Copy link
Member

@mattersoflight mattersoflight left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@edyoshikun I suggest that you improve the organization of files and documentation of demos (DL@MBL exercise, inference demos, instructions on how to run training with data/config/scripts we share) first. We can then iterate on the exercise.

ziw-liu and others added 3 commits August 7, 2024 16:39
* first draft of contrastive learning model

* fixed stem and projection head, drafted lightning module

* Contrastive_dataloader (#99)

* initial dataloader.py

* Update dataloader_test.py

* Update dataloader_test.py

* Update dataloader_test.py

* Update dataloader_test.py

* rename training script

* move contrastive network to viscy.representation module

* Update hcs.py

* refactored class names

* correct imports

* cleaner names for model arch and module

* new imports

* Fixed epoch loss logging and WandB integration in ContrastiveModule

* updated training_script.py

* Update hcs.py

* contrastive.py

* engine.py

* script to test data i/o speed from different filesystems

* moved applications folder to viscy.applications so that pip install -e . works.

* add resnet50 to ContrastiveEncoder

* rename training_script.py to training_script_resnet.py

* test dataloader on lustre and vast

* move training_script_resnet to viscy.applications so that `pip install -e .` works

* refined the tests for contrastive dataloader

* sbatch script for dataloader

* delete redundant module

* nits: updated the model construction of contrastive resnet encoder.

* Updated training script, HCS data handling, engine, and contrastive representation

* Fix normalization, visualization issues, logging and multi-channel prediction

* updated training and prediction

* update training and prediction script

* formatting

* combine the application directories

* lint

* replace notebook with script

* format script

* rename scripts conflicting with pytest

* lint application scripts

* do not filter all warnings

* log instead of print

* split data modules by task

* clean up imports

* update typing

* use pathlib

* remove redundant file

* updated predict.py

* better typing

* wip: triplet dataset

* avoid forward ref
this might increase code analysis time a tiny bit
but should not have any effect at runtime

* check that z range is valid
and fix indexing

* clean up and explain random sampling

* sample dict instead of tuple and include track index

* take out generic HCS methods for reuse

* implement TripletDataModule

* use new batch type in engine

* better typing

* read normalization metadata

* docstring for data module

* drop normalization metadata after transformation

* remove unused import

* fix initial crop size

* Infection state (#118)

* updated prediction code

* updated predict code

* updated code

* fixed the stem and forward pass (#115)

* fixed the stem and forward pass

* update forward calls to encoder

* self.encoder -> self.model

* nits

* l2 normalize projections

* black compliance

* black compliance

* WIP: Save progress before merging

* updated contrastive.py

* stem update

* updated predict code

* Delete viscy/applications/contrastive_phenotyping/PCA.ipynb

* pushing dataloader test updated

* pca deleted

* training and dataloader test

* updated structure

* deleted files

* updated training merged files

* removed commented code

* removed uneeded code

* removed uneeded code

* removed comments

* snake_case

* fixed CI issues

* removed num_fovs

---------

Co-authored-by: Shalin Mehta <[email protected]>

---------

Co-authored-by: Shalin Mehta <[email protected]>
Co-authored-by: Alishba Imran <[email protected]>
Co-authored-by: Alishba Imran <[email protected]>
Co-authored-by: Alishba Imran <[email protected]>
Co-authored-by: Duo Peng <[email protected]>
* wip: use lightning's tensorboard logger instead of wandb

* private logging methods

* log center slice only

* fix tensor cloning

* only log metrics on epoch

* add simple demo training script

* fix flaky test

* log graph + profiling

* switch to simple profiler

---------

Co-authored-by: Shalin Mehta <[email protected]>
@edyoshikun
Copy link
Contributor Author

@ziw-liu merging this.

@edyoshikun edyoshikun merged commit baa4ee3 into main Aug 8, 2024
4 checks passed
@mattersoflight mattersoflight deleted the dlmbl_2024_exercise branch August 28, 2024 15:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
documentation Improvements or additions to documentation
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants