Transforming images to look like paintings

Part 1 (2019)
1

I have finished part 1. And in lesson 7, Jeremy talk about the possibilities on GAN’s and he mentions making photos look like Monet paintings. I thought this would be a fun project, so i decided to be slightly original, and make something that makes paintings look like Van Gogh paintings.

I am using this small dataset: https://figshare.com/articles/From_Impressionism_to_Expressionism_Automatically_Identifying_Van_Gogh_s_Paintings/3370627

So the idea is to make something that can take any picture and make it look like it is a Van Gogh painting.

The only thing is that I am getting a little bit stuck here in the beginning. I would actually like to use feature loss, but I don’t think i can with this kind of problem. I don’t really have a Van Gogh and not Van Gogh version of my pictures.

Could i get a few pointer as to what would be a good starting point?

1 Like
2

I would start by look at the 2018 Lesson 13: https://youtu.be/xXXiC4YRGrQ?t=3688

I think the topic was covered very well in 2018 and I look forward to see how it will be covered in 2020.

3

Sylvain did an awesome blog post on this topic: https://sgugger.github.io/deep-painterly-harmonization.html#deep-painterly-harmonization

His notebook is here: https://github.com/sgugger/Deep-Learning/blob/master/DeepPainterlyHarmonization.ipynb

4

Neural Style Transfer is what you are looking for.
We extensively treated it in this thread Fast Style Transfer in fastai v1

4 Likes
6

Thank you so much guys. I was hoping there was a later lecture that touched on it. Thank you for the resources and thank you for the thread link.