Github’s Junyanz (UC Berkeley pixel researcher Jun-Yan Zhu) has this eye-popping routine for turning images of things into images of other things. like , or , or even .
I shared this before here: A Neural Algorithm of Artistic Style, which is a good thread to read. But I suppose this deserves its own thread.
I learned about it this morning while poking around Interactive Colorization on GitHub. I could see training it to turn landscape photos into architectural photos and vice versa, and the documentation seems to explain how that can be done. I’d suggest AI should be a tag available when posting?
Yes, clearly the topic of AI is of interest to many users and visitors. My top links tend to fall under this category. Maybe call it Computational Photography? Machine Learning?
I’d prefer that we use tags rather than have a ton of subforums. If we really have enough posts to warrent a sub category, I’d be happy to move all of them
Right. I generally place them in Lounge as a discussion point but adding a tag would be a good idea.
I recall this article from several years ago. May be of interest to you.
Also, did a simple internet search and ran into this program; haven’t vetted so be careful.
It seems to use the pix2pix model which is included in the CycleGAN package. I tried the CycleGAN trainer yesterday. It seemed to work, although the output was garbage as I only let one training epoch go by. It defaults to 200 iterations which I’m sure a big GPU could rip thru.
CycleGAN turning a photo of a stream near Mt Wynne, California into a PC Board