Koopa Luath's test/demo releases

After looking through the Doom9 threads, I've decided that when I upscale, at no point in the process is PointResize going to be involved. It'll just make the quality look worse than if I just used NNEDI and/or EEDI.

[spoRv]

...at no point in the process is PointResize going to be involved.

Well, I think it was clear after I looked at resize algorithms, maybe two minutes after that!

[CSchmidlapp]

After looking through the Doom9 threads, I've decided that when I upscale, at no point in the process is PointResize going to be involved. It'll just make the quality look worse than if I just used NNEDI and/or EEDI.

http://www.infognition.com/articles/what...ution.html

As for NNEDI and EEDI, I think using them in concert with each other would be better than just using one or the other exclusively.

[Colek]

https://github.com/mysteryx93/AviSynthShader

I found this to give best results.

Cool! Can you show me a quick example as you might've used it?

[Colek]

SuperResXBR(2, Str=1, Soft=0.2, XbrStr=3, XbrSharp=1.5, fWidth=1920, fHeight=1080, PlanarOut=false)

[CSchmidlapp]

https://github.com/mysteryx93/AviSynthShader

I found this to give best results.

Does this work better than Super Resolution linked above?

https://github.com/mysteryx93/AviSynthShader

I found this to give best results.

Does this work better than Super Resolution linked above?

It is the same super resolution (well very similar) but in Avisynth form with many improvements like XBR scaling (in the lastest version) and tweakable settings which info's version lacks.

It's better.

[TomArrow]

From what I understand they have only the name in common ("Super Resolution")

They are based on different concepts - the infognition one uses a temporal approach combining details from neighboring frames - whereas the other one uses some unique method of upscaling then downscaling and comparing to the original or whatever

Either way - the infognition one also has an AVIsynth filter for cheap money - I have it

Maybe you can even combine them - as the latter accepts a scale algorithm as input I think - and see if you can maximize results