Labkit improves support for segmentation ground truth editing. (preview)

Dear image scientists,

Some people already use Labkit to create segmentation ground truth, for the training of machine learning tools like StarDist, DenoiSeg and similar tools… Now, this works even better!

Labkit has new features that makes it simpler to manually create segmentations. A preview version is available from this update site:
https://sites.imagej.net/Labkit-Preview/

Improved brush tool

The brush tool now has super stable graphics. The flickering and low resolution rendering that sometimes occurred is gone.

Better handling of many labels

  • Opening a labeling with many labels is faster.

  • The label list always shows the selected label.

  • There’s a color picker tool for easy label selection.

The “override” flag…

…in the drawing toolbox, prevents you from accidentally creating overlapping labels.

(Disclaimer, Why is this a preview:

  • The update site includes a preview version of Big Data Viewer. It is therefore very likely incompatible with other Big Data Viewer based tools like Big Stitcher, Mastodon, etc.

  • The pixel classification algorithm shipped with this version is undergoing changes. Future Labkit versions might not be able to open classifiers trained with this version of Labkit. Please use the usual update site https://sites.imagej.net/Labkit/, if you want to do automatic segmentation.)

So if you want to manually create segmentations, try the Labkit preview update site:
https://sites.imagej.net/Labkit-Preview/

9 Likes