Superresolution vs Resize - worth it for realtime?

[4evrplan]

Is there any noticeable difference in quality between superresolution and a good quality resizer like Lanczos for video?

Seems like at 30+fps, the human vision system may not notice any difference at all, or that the brain may in fact may be subjectively combining detail from frames anyway so that you get the perception of more detail even without the computer doing the work.

The reason I ask is that I'd like to play back my SD sources (like DVD) on my 1440x900 monitor with the best possible perceptual visual quality my computer can manage.

Which leads to another question: Are there shaders that do this type of superresolution/scaling/sharpening on the GPU or will I have to use Avisynth for that? Also, what's the latest/best avisynth script and/or shader for this task?

[CruNcher]

We already have a lot of discussion threads going on on this matter (with different solutions and comparisons) in the Avisynth forum, you could revive 1 of them :)

http://forum.doom9.org/showthread.php?t=154141
http://forum.doom9.org/showthread.php?t=153651
http://forum.doom9.org/showthread.php?t=100114
http://forum.doom9.org/showthread.php?t=147695
http://forum.doom9.org/showthread.php?t=160732
http://forum.doom9.org/showthread.php?t=113391
http://forum.doom9.org/showthread.php?t=142704

[4evrplan]

No need to reopen them (funny I started one of those threads, but it was closed for being OT), scanning over those threads seems to answer my first question pretty conclusively; super-resolution isn't worth it.

That just leaves the second question, and the reason I posted here instead of the Avisynth forum: are there any shaders I could use with a supporting player, say potplayer, that do very high quality upscaling, or am I better off with nnedi3_rpow2 or spline64resize. I could see this being really nice and fast on a decent GPU.

[CruNcher]

scaling shaders are more complex they need to be implemented @ the core else you would need to always define a resolution to scale too ;)
Some new scaling shaders http://www.mediafire.com/?025vsef8ads6t5x are implemented in Jans Experimental Renderer you could combine those with something like Sharpen Complex 2 to get a sharper output in the End depending on the Input it will loadup the GPU though, there is also MadVR where you could combine shaders and later do additionally sharpen via for example ffdshow (CPU) or sharpen via the GPU (also on the shader) for example via fft3dgpu :)
Though most probably the Hardware scaling algorithm these days are very efficient Intels SB introduced scaling for example is a adaptive lanczos4 based scaler (supports blending) so more efficient then mitchell netravalis or bicubic which was used in the past the most in GFX Hardware :)
It seems to be more efficient then what Nvidia and AMD have currently implemented, currently though it is restricted to EVR, ffdshow-quicksync will get it implemented someday though it can be utilized already via the IMSDK ;)

http://forum.doom9.org/showpost.php?p=1552011&postcount=572
http://forum.doom9.org/showpost.php?p=1552089&postcount=577

nnedi3 is still the highest quality (there aren't any compares with Intels implementation yet) though as everyone knows its also the slowest and the perceptible quality/speed ratio might be not so good @ all for everyone ;)