Experimenting with AI upscaling and making children cry.

I needed some really terrible quality old video to try my hand at machine-learning upscaling. This was about the worst I could find. This program was made to traumatise six-year-olds into paying attention in school. The results still look pretty crap, mostly because of the unnaturally smooth faces. I was able to largely solve temporal stability problems with a little pre-filtering, but really this is as good as it's going to get without a recurrent network. The original was just too far gone: There was no detail in the faces at all for the upscaler to work with. It did a good job on backgrounds though.