keras network learner does not see image column

[l.thomas]

Dear all,
I came across a puzzling issue, I have a functional workflow using the keras network learner node for image-classification similar to the example workflow with cats and dogs.

However, on another machine, with the same data, workflow, and as far as I know same configuration, the keras network learner node does not see the image column !
The configuration window only shows the option for integer or double.
I’m quite sure I had the same issue with the Cats and dogs example, so I don’t think that it’s an issue with the workflow or data (224x224x3, FloatType images).

Where it works

On the other machine

Both configurations run

• Knime 4.1.3
• 
  
  and the CPU Deep learning environment as created by the KNIME Deep Learning preferences
  (Python 3.6.10, TF 1.12.0 and Keras 2.2.4)

[Kathrin]

Hi @l.thomas,

could it be that you don’t have the following extension installed on your second machine?

KNIME Image Processing - Deep Learning Extension

Best,
Kathrin

[l.thomas]

Oh yeah that was it, thanks a lot !
I am so used to having the extensions being installed automatically that I did not spotted it.

Is there a way to manually mention this extension as a “dependency” of the workflow, so that it is installed when downloading the workflow from the hub for instance ?

[Kathrin]

Not that I know, but this is a great idea! I will ask the developers, whether this is somehow possible.