Tutorial on how it works
SelaO opened this issue · comments
Is there a tutorial or guide on how this works and how you developed it?
If not, I really think you should make one for beginners.
It's a neat project that can really motivate people to learn about ML.
It work by Generative adversarial networks (GANs) where you can find numerous resources online.
However, GANs are notoriously difficult to train, especially for this large network; there exists hundreds of choices of existing NNs and parameters that may not be explained.
But in short, there are some well defined discriminators inside.
A model is trained by by a sketch set and a painted set. parts of results are generated by U-Nets.
It is not a trivial network but for a beginner.
You can take https://github.com/pfnet/PaintsChainer as a reference and it includes model training code inside for you to learn (a lot).
The network structure is different from style2paints though.
Is there any course or tutorial that you can recommend?
@SelaO You can starts from how CNN works? and then how GAN works? and then I suggest a little content of GAN from https://github.com/soumith/ganhacks for you to start.
here is a full example of using DCGAN
https://github.com/jacobgil/keras-dcgan
but it is not enough for you to GAN some real Anime images.
vgg, inception, resnet are very good models to discriminate or classify images. You can have a try.