VisCy 0.3.0rc1

Warning

This is a release candidate for testing.

VisCy 0.3.0 incorporates the representation learning task as a core feature.

Breaking change

The top-level CLI now supports both image translation and representation learning tasks. This required changes to configuration files. Concretely, data and model fields now require import paths to be specified. See the updated examples for reference on migrating existing virtual staining configs.

What's Changed

• Add script to visualize effective receptive field by @ziw-liu in #144
• adding VS hugginface demo by @edyoshikun in #172
• Single-cell representation learning by @ziw-liu in #153
• Vendor pad shape function by @ziw-liu in #189
• Fix module name spelling by @ziw-liu in #190
• Add new author to citation by @ziw-liu in #188
• Add links to hosted files and napari-iohub wiki by @ziw-liu in #192
• Bump MONAI to unpin NumPy by @ziw-liu in #194
• Add badges to readme by @ziw-liu in #197
• Simplify development installation by @ziw-liu in #198
• Fix validation loss aggregation in VSUNet by @ziw-liu in #202
• Expose prefetch_factor and persistent_worker for the HCS datamodule by @edyoshikun in #203

Full Changelog: v0.2.1...v0.3.0rc1

VisCy 0.3.0rc0

Warning

This is a release candidate for testing.

VisCy 0.3.0 incorporates the representation learning task as a core feature.

Breaking change

The top-level CLI now supports both image translation and representation learning tasks. This required changes to configuration files. Concretely, data and model fields now require import paths to be specified. See the updated examples for reference on migrating existing virtual staining configs.

What's Changed

• Add script to visualize effective receptive field by @ziw-liu in #144
• adding VS hugginface demo by @edyoshikun in #172
• Single-cell representation learning by @ziw-liu in #153
• Vendor pad shape function by @ziw-liu in #189
• Fix module name spelling by @ziw-liu in #190
• Add new author to citation by @ziw-liu in #188
• Add links to hosted files and napari-iohub wiki by @ziw-liu in #192

Full Changelog: v0.2.1...v0.3.0rc0

v0.2.1

Patch release to update README and example notebooks.

What's Changed

• version lighting CLI example by @mattersoflight in #128
• Updated code (contrastive learning) by @alishbaimran in #130
• Configurable drop path rate in contrastive models by @ziw-liu in #131
• Config-based prediction with Xarray-based output format by @ziw-liu in #132
• Plot tracks in latent space and real space by @mattersoflight in #135
• Fix deprecated custom forward method by @ziw-liu in #151
• updating the notebook after running it at DLMBL2024 by @edyoshikun in #149

Full Changelog: v0.2.0...v0.2.1

VisCy 0.2.0

VisCy 0.2.0 adds the following features:

• Application scripts for single-cell infection classification through semantic segmentation
• Tutorial notebook that demonstrates the virtual staining pipeline
• Test time augmentations in the virtual staining prediction writer
• (Alpha) Experimental support for single-cell phenotyping through contrastive learning

This release maintains compatibility with the virtual staining model weights from the v0.1.0 release (download link).

What's Changed

• Update dataset URL for demos by @ziw-liu in #103
• Bump lightning and matplotlib by @ziw-liu in #105
• Cellular infection phenotyping using annotated viral sensor data & label-free images by @Soorya19Pradeep in #70
• Pin numpy due to MONAI bug by @ziw-liu in #111
• Updating demo notebook for training by @edyoshikun in #100
• Test time augmentations by @edyoshikun in #91
• Update demo setup script by @ziw-liu in #112
• Single-cell phenotyping with contrastive learning by @ziw-liu in #113
• Migrate from wandb to tensorboard by @ziw-liu in #122
• Adding link to demos and library of VS models wiki by @edyoshikun in #119
• Tune augmentations with CLI and config for contrastive models by @ziw-liu in #126
• DLMBL 2024 notebook by @edyoshikun in #114

New Contributors

• @Soorya19Pradeep made their first contribution in #70

Full Changelog: v0.1.1...v0.2.0

VisCy 0.2.0rc0

VisCy 0.2.0 adds the following features:

• Application scripts for single-cell infection classification through semantic segmentation
• Tutorial notebook that demonstrates the virtual staining pipeline
• Test time augmentations in the virtual staining prediction writer

This release maintains compatibility with the virtual staining model weights from the v0.1.0 release (download link).

What's Changed

• Update dataset URL for demos by @ziw-liu in #103
• Bump lightning and matplotlib by @ziw-liu in #105
• Cellular infection phenotyping using annotated viral sensor data & label-free images by @Soorya19Pradeep in #70
• Pin numpy due to MONAI bug by @ziw-liu in #111
• Updating demo notebook for training by @edyoshikun in #100
• Test time augmentations by @edyoshikun in #91

New Contributors

• @Soorya19Pradeep made their first contribution in #70

Full Changelog: v0.1.1...v0.2.0rc0

VisCy 0.1.1

Patch release to update the README.

What's Changed

• README update by @ziw-liu in #96

Full Changelog: v0.1.0...v0.1.1

VisCy 0.1.0

This is first release of VisCy, the machine learning pipeline to train and deploy computer vision models for single-cell phenotyping.

With 0.1.0 the following key features are available:

• Training, evaluation, inference, and deployment of virtual staining models based on 2D Residual U-Net, 2.5D U-Net, 3D U-Net, and UNeXt2 architectures
• Data module implementations for HCS OME-Zarr datasets, as well public test datasets like LiveCell and CTMC v1.
• Composing datasets and transformations for training and validation
• Distributed (DDP) training

The weights of the virtual staining models reported in the preprint can be found in the binaries section below.

What's Changed

• Migration from microDL by @ziw-liu in #1
• Update README by @ziw-liu in #22
• Bump iohub version by @ziw-liu in #25
• readme + dependencies tested with python 3.10 by @mattersoflight in #30
• Demo notebooks by @ziw-liu in #29
• 3D augmentation and 2.1D U-Net by @ziw-liu in #27
• Fix datamodule by @ziw-liu in #28
• Improve augmentation by @ziw-liu in #31
• Fix inference by @ziw-liu in #32
• viscy -> VisCy by @mattersoflight in #34
• Update readme.md typo for cd VisCy by @edyoshikun in #41
• 2.1D upscale decoder by @ziw-liu in #37
• dlmbl 2023 archive by @mattersoflight in #44
• Fix center slice metrics for 3D output by @ziw-liu in #51
• Configure the number of image samples logged at each epoch and batch by @ziw-liu in #49
• Example workflow by @ziw-liu in #45
• Project icon by @ziw-liu in #38
• Fix predicting new channels in an existing store by @ziw-liu in #57
• Document data methods by @ziw-liu in #50
• Visualize feature maps by @ziw-liu in #53
• Baseline 3D-LUNeXt by @ziw-liu in #58
• Preprocess CLI and source scaling during prediction by @ziw-liu in #59
• Configurable augmentations by @ziw-liu in #61
• Bump dependencies and update documentation by @ziw-liu in #64
• checkpoint as a model config parameter for warmup cosine learning rates by @edyoshikun in #66
• Masked autoencoder pre-training for virtual staining models by @ziw-liu in #67
• Filter empty detections in labels by @ziw-liu in #74
• Add CITATION.cff by @ziw-liu in #79
• Fix 3D to 2D prediction with UNeXt2 model by @ziw-liu in #80
• 2D FCMAE by @ziw-liu in #71
• Rename UNeXt2 by @ziw-liu in #84
• Test on Python 3.12 by @ziw-liu in #88
• Add preprint reference to README by @ziw-liu in #85
• Fix architecture name in network diagram script by @ziw-liu in #86
• Add the scale metadata to the output_stores by @edyoshikun in #89
• bumping to cellpose 3 by @edyoshikun in #92
• Scale metadata handling for positions by @edyoshikun in #93
• Demo for VSCyto2D and VSCyto3D by @edyoshikun in #94
• Fix demos on other platforms by @ziw-liu in #95

Full Changelog: https://github.com/mehta-lab/VisCy/commits/v0.1.0