I’m preparing images for a Keras Network. I used the Rotation 2D Node to generate “more Data”.
Does the Image Resize Node and the Keras Network work on the real data (stored on the harddrive) or are both Nodes able to use the virtual data in the Knime Table?
If the nodes need real data, how is it possible to duplicate the files?
Right now I’m not sure…
That’s a very cryptic question (or my brain is just not able to properly process it). Are you asking if the Keras Network Learner uses the images that you have generated (by manipulation images) or the pixel data from the images that are residing on disk? KNIME passes the derived images to the network for training (also for predicting btw) instead of the original images.
It doesn’t. But you could use an Image Writer to write the scaled and rotated image to TIFF files, for instance.
The pciture in Row11 is stored on the harddrive.
The picture in Row12 was the rotated one.
I combinded the two columns (thats why Row11 and then Row12).
What I now want to do is:
I have the original file and the processed one, both in a knime-table, but I dont want to use the image writer node because the processing is done within a workflow.
Are there temporary values for the images within the knime-table?
Is this Image[name=download233…bin;…]
A temporary copy of the picture which is shown in the thumbnail?
Note: the workflow is processed on a knime server
it really looks like, that this bin-file is the image of the knime-table, ist ist somehow possible to make it useable to generate the data on the server without using the local machine?
found out you can just use the image-string.
Yes and no. The image is still in memory until is cached to disk. So it might at some point end up on disk.
In any case, the Keras Network Learner will be fed with images from the table. That is, it shouldn’t matter if you have loaded them from disk or if they have been processed somehow.
Very likely I haven’t understood yet what the issue is… Sorry!