For information, I’m currently trying to write some code for the forward evaluation of such neural networks, and my latest tests show that the evaluation itself is probably fast enough to make it usable in G’MIC.
I’m not speaking about the learning phase though