Zenodo First Release

First official release of the empanada-napari code on Zenodo.

v0.2.1

New Features

• Added "Apply in 3D" option to proofreading tools. This enables faster proofreading of stacks of 2D images and flipbooks.
• Added merging from line and path shapes.
• Metadata generated for instance segmentations (like number of voxels and bounding boxes) are now attached for model predictions.
• Added tiled 2D inference for processing very large 2D images and "viewport" inference for testing models on small ROIs of much larger images.
• More operations now support multiscale datasets with scale and translation metadata.

Bug Fixes

• Fixed bug that prevent export of models trained and fine-tuned on CPU.
• Fixed bug in the "Use points as markers" split option that caused inconsistent results when applied to 2D images translated in 3D.

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

v0.2.0

What's Changed

• Finetuning support by @conradry in #12
  Full Changelog: v0.1.4...v0.2.0

New tutorials and detailed usage notes at: https://empanada.readthedocs.io/en/latest/empanada-napari.html

New Features

• Added the MitoNet_v1_mini model which uses BiFPN modules for parameter efficiency.
• Added support to easily pick training data from an individual or stack of 2D images and 3D volumes (isotropic and anisotropic). Allows manual selection of ROIs from dropped points.
• Added module to store annotated training data in the directory format expected by empanada.
• Added support for fine-tuning any existing models.
• Added module to print out descriptions of any registered models to aid in the preparation of fine-tuning datasets.
• Added support for training panoptic segmentation models from scratch or from CEM1.5M pre-trained weights.
• Updated the Register new model module to allow for remote (URL) model files. Switched from requiring a zip file to model config and paths to torch scripted models.
• Modified model config files to include a FINETUNE section with relevant (hyper)parameters.

v0.1.4

Bug Fixes

• Fixed SIGABRT during multi-gpu inference.

New Features

• More stable multi-gpu inference.
• Improved integration with the empanada library
• Default instance splitting uses distance watershed instead of marker watershed
• Jump to labels module to move slider to first slice where a particular object id appears
• Improved postprocessing of semantic and instance segmentation classes after panoptic segmentation
• Added option to choose inference axis in 3D (i.e., xy, xz, or yz)

v0.1.3

Bug Fixes

• Hard-coded merge_iou_thr and merge_ioa_thr. Removed from the update_params method on Engine3d.

New Features

• Added support for running 2d inference in batch mode on a stack of images (optionally loaded from a folder).
• Converted to npe2 with napari.yaml manifest.

v0.1.1

Stable napari plugin

v0.1.0

First public release of the plugin

Initial plugin version

All basic plugin functionalities are implemented. Model testing on arbitrary axes and slices, 2D stack inference and 3D orthoplane inference. Not finished are options for storing results are OME-ZARR or keeping results loaded in memory as bumpy arrays.