hi forum,
I followed this tutorial (https://www.linkedin.com/pulse/learn-how-classify-images-more-than-2-classes-using-knime-purnomo-cnvic/) and built my own workflow based on the workflow show in this article:
I’m trying to classify images of toolboxes from a manufacturer, where the images could show the toolboxes in total with open lid empty, closed lid or with open lid plus decoration objects and detail views of features (compartments, lock, etc)
I have set up my workflow, just the same as in the tutorial, however, with four classes instead of three. Training went well and the execute run as well, but yielded only one class for all my images (only one of the dense_2-columns show 1 all others 0).
Where would I have to look first to adjust my settings?
the distribution of classes look like this (training data):
is it being the sample too small? too unevenly distributed?