3 ms·
Very beautiful. If I trained the network on say, just drawings and painting Mondrian did, do you think they'd converge stylistically? Is a huge (5,000+ items)
by sova 5y ago
Very beautiful. If I trained the network on say, just drawings and painting Mondrian did, do you think they'd converge stylistically?
Is a huge (5,000+ items) dataset required to get decent results?
- mfi 5y agoThanks! I think it's quite likely that you will heavily overfit the network if you only use a handful of images, where it might end up reconstructing the exact paintings. I only have experience using 5,000+ images.
- TheRealNGenius 5y agoReally cool that you brought up Mondrian, since I literally implemented a Haskell program that generated art in their signature style during college. No AI required at all, just setting some random parameters. Here are some results: https://imgur.com/a/rtWkTTl https://imgur.com/a/rtWkTTl
- sova 5y agoEverything was great until I saw a curved line... Honestly though much better than a lot of people who try to imitate Mondrian