site banner

Small-Scale Question Sunday for October 30, 2022

Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?

This is your opportunity to ask questions. No question too simple or too silly.

Culture war topics are accepted, and proposals for a better intro post are appreciated.

4
Jump in the discussion.

No email address required.

With all that progress in machine learning, WHEN will be we able to upscale frame rate of old animes from effective 5-8 fps to 60 fps?

Early pre-neural network attempt (SVP) is about as effective as nil, older DAIN still leaves a lot to be desired.

It should doable in 2-3 years with ease, if someone cared enough.

The technology might already be there. There simply isn't enough money to want to do it.

There is also the fact that some of it might be impossible. The physics of an anime only make sense in 5-6 fps. If you try to scale it up, a human might be just as unable to do it as a machine.

Most ML models today match the absolute best human beings, because they are ones who label it. So, if a human can't do it, a machine will surely be unable to do it.

But, it should be totally possible to build something that looks plausible without artefacts for anime that has physical consistency to it.

So, if a human can't do it, a machine will surely be unable to do it.

I'm fairly sure there's humans who can do it, just it takes a lot for time to learn to draw and even still it's labour-intensive.

This is actually a case where a fan project is the probably the best bet.

There are some unique issues. Different elements update at different frame rates. Also they will scroll at full frame rate while animating at a lower one.

But there's not enough commercial or academic interest to pump money into research.

Fans and animation enthusiasts can put together great training sets on their own. There are enough nerdy types to play around with models.

Different elements update at different frame rates.

Yes, correct, but it seems to be a little easier task than deducing correct animating of body parts from known 4 running frames.

I heard there's Google frame rate upscaler, thought it seems it has no specific support for anime...