Visualize predicted labels on image

[apoliti]

Dear Knime,

I wonder if it is possible to see the results of an object classification like the classification done in NucleiClassificationTraining (the workflow is a little old but one can do something similar) in an image.
Basically something like the manual label editor, but where new labels are set by a predicted class.

Nuclei Classification Trainig – manuel

Thanks

antonio

[apoliti]

I tried following workflow. What I miss is something that allows to evaluate the predictions. Best is a new labeled image.

[apoliti]

Hi @gab1one,

I just ping you as maybe there is a quick way to do what I want to do.

Thanks

Antonio

[apoliti]

Add an additional person to comment on this feature @stelfrich

[apoliti]

Hello @imagejan,
I think you do work with Knime quite a bit.
Do you have a solution for my question?

Antonio

[imagejan]

Hi @apoliti, and sorry for the delay!

May I introduce you to the great powers of the GroupBy node, combined with the aggregation functions added by KNIME Image Processing (KNIP)?

KNIP adds several aggregation functions to aggregate images from multiple rows:

• Compose Image – create a new image of the same type, with the single input images places at the position of their known ‘offset’ (you can use Set Image Metadata to manipulate them),
• Compose Labeling – create a new segmentation image from input images and labels,
• Merge Image – create a new image (stack) with a N+1 dimensions from N-dimensional input images,

and some more that are equivalent to those in Projector, but working on multiple rows instead of single (stack) images:

• Max Image
• Mean Image
• Min Image

---

Here’s an example workflow that illustrates how to create a label image from given segments and labels:

Create Labeling from Bitmask and Labels – imagejan

Note that the Edit button in the Parameter column of the Manual aggregation tab contains all the magic:

Hope that helps, and happy image processing !