sxela / face2comics

face2comics datasets

face2comics by Sxela (Alex Spirin)

This is a paired face to comics dataset, which can be used to train pix2pix or similar networks.

This dataset contains a lot of crappy nightmare-fuelish samples, which tend to be useful for full image to comic convertion, as they teach models how to trace something if it's not a face.

Update: v2.0.0 dataset added!

v2.0.0 dataset

Darkish red style, 1024x1024, 10000 pairs (total 20k images) Download faces from github (~1Gb) Download comics from github (~1Gb)

2x2 dataset sample:

10x10 dataet sample:

An inference sample of a fastai unet trained on this very dataset :

You can train your own unet by using their awesome notebook here

v1.0.0 dataset

Dark style, 512x512, 10000 pairs (total 20k images) Download from github (~700Mb)

2x2 dataset sample:

10x10 dataet sample:

An inference sample of a fastai unet trained on this very dataset :

You can train your own unet by using their awesome notebook here

Recognition

justinpinkney for his layer blending idea, repo and tutorials https://github.com/justinpinkney/stylegan2

NVlabs for their Stylegan2, of course, though I didn't use their original code https://github.com/NVlabs/stylegan2

fastai for their awesome framework and simple GAN training pipeline https://fast.ai

face2comics's Issues

Source Code for Fast Ai Unet

Hello,
Can you please provide source code for Fast AI UNet trained on dataset you provided?
Thank you.

v2.0 original dataset

Hi! It’s really great work! I’m learning GANs and try to run unet on the paired dataset you provided, however the result is not as good.

Could you please provide the original dataset for the v2.0.0 so that I can play around with StyleGAN blending and make it work?

would be very appreciated!

How you mixed the ffhq model with comic faces?

Could you post the original dataset of the comic faces?

Thanks for you work! I want to experiment with stylegan2 blending. Can you post the original face dataset on which you finetuned stylegan?

I trained so poorly?

100 pairs of data sets with 1024X1024 resolution are trained according to 512, and the pixel2pixelHD inference is used. Why is the result so poor? Do you know why? The data set pairing adopted is here :https://github.com/Sxela/face2comics