sadfun Feb 27 2019 at 14:04

We're in UltraHD Morty! How to watch any movie in 4K

3 min

14K

Working with video*Machine learning*DIY

+17

Comments 7

kashey Feb 27 2019 at 14:51

So, 82 + 26 GB. But what's about the __time__? Probably this is not a quick operation.

sadfun Feb 27 2019 at 16:36

No, ffmpeg did each operation just in 35 minutes (MacBook Pro 2018, core i7)

kireevmp Feb 27 2019 at 16:39

Potentially, other enhancing methods (not Photoshop, I mean) make possible to build a cluster of parallel jobs to enhance all the images really fast, so it'll be like a great system of making low-res films beautiful in real time (or not real, but very fast)

sadfun Feb 27 2019 at 16:53

Cool idea, but neural multithreading could turn your PC into a fiery box :)

Although, something similar already exists: madvr.com has neural-based upscaling method, called NGU.

limassolsk Feb 27 2019 at 17:29

The question without answer from previous post:

Did you try to do the same by using ffmpeg only?
ffmpeg -i input.mp4 -vf scale=3840:-2 -c:v libx264 -preset slow -crf 22 -c:a copy output.mp4

sadfun Feb 27 2019 at 17:56

This is a regular scale, the result will have more pixels, but the same quality as the original.

kikiwora Feb 27 2019 at 18:30

That's definitely quite handy.
But question, is it possible to do upscale in real time?
The thing is I often don't have enough free space, and thus wouldn't be able to convert movies using this method. Is is interesting, but not really viable.

And, well, ofc if i want to download some movies to my disk, I'll need even more disk storage for 4K versions, but if the real-time upscale is possible using regular video players, this problem could be easily avoided.

Also, I would like to know how your method compares to the regular TV/TVBoxes 4K upscales?

I assume that is is possible to use something like VLC filter to upscale video in real-time, but I have no idea how reliable it is compared to this simple, yet working method.