Releases: TheJacksonLaboratory/activelearning
Improved sampled positions
This version improves how sampled positions are managed.
Additionally, this version fixes an issue of a high resolution mask (same as the input image) being loaded for patch sampling.
Improved layers management
This version improves the way how layers are managed and used in the active learning pipeline.
There is a new check box to select the layer group used as labels for the fine-tuning step, rather than using exclusively the generated segmentation layer.
Additionally, more models were added to the list of available pre-trained models for cellpose in its corresponding widget.
Improved compatibility with Cellpose
This release removes the upper version constraint on cellpose, and would work with existing >=3.0.0 versions of it (currently 3.1.0).
That allows users to install more recent versions of cellpose package.
Fixed multiscale data layers creation
This versions solves a problem caused by using tensorstore to create multiscale data layers in napari.
The temporary solution uses zarr to open the files directly.
Fixed data group parsing in zarr filenames
This release corrects issues of how the data group is parsed from zarr filenames.
The data group contains the pixel data that is used for segmentation and fine tuning processes.
Improved management of Label objects
This version fixes several bugs related to Label objects management when interacting between other object managers.
Additionally, a ROI is defined for labels and not only image data. Before this, labels wouldn't match their input images because of the difference in scale.
Labels navigation buttons enabled
This version enables navigation buttons for labels, that were disabled at initialization and never enabled afterwards.
More improvement to cellpose module for fine tuning
This version requires cellpose>3.0.0,<=3.0.10 directly as optional dependency, instead of napari-cellpose.
The pinned version exposes their train module that enables fine tuning their models.
Other available pretrained models have been added to the selectable list of models for segmentation and fine tuning.
Improved cellpose model selection for fine tuning
This release integrates an improved pyproject.toml file for package installation, as well as a fix to a bug that prevented the selection of different models for cellpose after initialization.
