looks great so far
#1
by
Zarxrax
- opened
This is very impressive. Do you think it could be improved by any significant amount with a better dataset or more training, or is this about as good as it can get with the current technology?
I spent a lot of time training, but I found no obvious improvement. I think we need a better manually labeled dataset rather than the current dataset that generating images and labels by background and foreground images.
I was thinking visual novels might make a good dataset for something like this, because they typically have character images that can be overlaid on different backgrounds.