Generation of new images from existent ones using deep learning

[njssnjss]

I would like to ask if anyone as built a Knime workflow to use deep learning (tensorflow/keras) using GAN’s or else… to generate new “creative” images from a training dataset of images?

Thank you.

Nelson

[nemad]

Hello @njssnjss,

we at KNIME haven’t attempted GAN’s in KNIME, yet.
This is mostly because the existing learner (Keras Network Learner) does not support the training strategy of GANs (i.e. alternating between training the generator and discriminator) and thus such a workflow would have to rely on the DL Python nodes.
Please let us know if you successfully build such a workflow and even more so if you run into roadblocks along the way, so we can see what functionality we need to add in future releases.

Best,

Adrian