to get a good and intuitive idea whether a new image similarity measure works well or not I`d like to cluster my images (many of them, lets assume 10k) by their respective similarity to each other - I can calculate all these cluster using various different approaches - my question is more on how to visualize the end result including the clustered images (while the actual clustering of course works on a feature vector). I could imagine to create a network where each node is represented by its image value - or anything else really that gives a quick visual idea whether the metric makes sense. Does KNIME support any of those approaches? I am also open to new suggestions :)
Do you plan to put the images themselves into your visualization? Like a force layout but each point is a small image? There is currently no implementation of that in KNIME. If you have a R / Python / Javascript (e.g. with D3.js) script that creates such a layout for images, you could use that within KNIME and also do your other processing and analysis steps there.
I found this blog post that shows how to create a force layout with custom node graphics with D3: https://bl.ocks.org/mbostock/950642. It could be a basis for your visualization.
If you have any more questions, feel free to ask them,