>>151009159
Well Daiz?
>>151099223
That wasn't Daiz you fucking retard.
>replying to fakes
See >>151020103
Because upscaling this one frame took an RX480 eighteen seconds.
daiz Taneli Vatanen
>>151099631
For comparison, here's something I threw together super quick with some Avisynth filters that runs over 50 times faster.
>>151100418
How does a common man upscale images with this?
>>151100418
How does the NGU compare to NNEDI3?
>>151100418
Look at that fucking banding jesus fuck.
>>151100582
Haven't compared them myself so can't say for sure.
>>151100657
It's a quick example made from CR video, what do you expect. JPG compression doesn't really help either but PNGs at 2160p are just too big to post here.
>>151100418
The differences are quite visible from up close.
From one side the waifu2x doesn't have visible compression errors, on the other side it washes out colors and blends together stuff that shouldn't.
Then again there are filters for MPC that remove compression errors in real time, but I'm not sure how much CPU intensive that would be with a 4k resolution.
>>151100987
>From one side the waifu2x doesn't have visible compression errors
This is because of the super-aggressive denoising that waifu2x has. It certainly looks nice on frames consisting largely of flat areas like this but it completely obliterates things like textures elsewhere. I could've used as aggressive denoising for the comparison myself but honestly it's just cheating because it's not something you should use when actually processing video. Picture related.
>>151099631
>>151100418
While I agree the Avusynth one looks better, some of the filters used don't really know ho to handle image borders.
>>151099223
>waifu is a meme