Why would one want to run machine learning inference from #java?

To do so on 3D, 4D, ND datasets, trivially accessible from image processing and visualization libraries such as #ImgLib2, the #BigDataViewer, #LabKit and more, all integral parts of #FijiSc.

* LabKit: https://imagej.net/plugins/labkit/

* BigDataViewer: https://imagej.net/plugins/bdv/

* ImgLib2: https://imagej.net/libs/imglib2/

* Fiji: https://fiji.sc

Now onto #FijiSc: Fiji is a recursive acronym meaning "Fiji is just ImageJ" https://fji.sc (and the paper https://www.nature.com/articles/nmeth.2019 ) –and #ImageJ is a #java open source software for image processing https://imagej.nih.gov/ij/index.html written by Wayne Rasband from the #NIH Research Branch.

An analogy: think of ImageJ as the kernel and Fiji as the rest of the operating system.

#FijiSc brings to #ImageJ:
(1) a package manager to install and update plugins, and that crucially enables reproducible science by exporting the whole set of plugins and libraries as an executable;
(2) a Script Editor https://imagej.net/scripting/script-editor supporting many languages (#python, #groovy #ruby #scala #clojure and more), all with access to a huge collection of #JVM libraries;
(3) huge amount of libraries such as #ImgLib2, #JFreeChart for plotting, for GUIs, etc.

There are many, many plugins. A tiny sample:

Machine learning-based image segmentation:
- #LabKit https://imagej.net/plugins/labkit/
- #WEKA Trainable Segmentation https://imagej.net/plugins/tws/index

3D/4D/ND Visualization:
- 3D/4D Viewer #3DViewer https://imagej.net/plugins/3d-viewer/index with ray-tracing, orthoslices, volume rendering, and more
- #BigDataViewer #BDV https://imagej.net/plugins/bdv/index for interactively navigate N-dimensional image volumes larger than RAM

Image registration and serial section alignment:
- #BigStitcher for registering 3D/4D tiled datasets, with multiview deconvolution and more https://imagej.net/plugins/bigstitcher/index
- #TrakEM2 for montaging in 2D and alinging in 3D collections of serial sections, typically from #vEM (volume electron microscopy) https://syn.mrc-lmb.cam.ac.uk/acardona/INI-2008-2011/trakem2.html
- #mpicbg libraries for extracting #SIFT and #MOPS features, then finding feature correspondences and estimating rigid and elastic transformation models https://www.nature.com/articles/nmeth.2072

Summarizing #FijiSc is impossible. See the online forum where questions find answers by the hand of the broader community of users and developers https://forum.image.sc/