[REQ] SmoothD2 for VapourSynth

[Boulder]

Jim Conklin's nice deblocking filter for Avisynth cannot be used with the x64 version of VapourSynth as it's compiled with Avisynth 2.5 headers. If there's someone bored out of his/her mind, I'd be grateful if the filter was ported to VapourSynth :)

https://sites.google.com/site/jconklin754smoothd2/download

[Myrsloik]

You forgot to mention that there isn't an x64 Avisynth version. And the reason for that is piles and piles of inline asm. Also piles of dead code in this. And of course it uses a custom memcpy.

A+++++

Would Avisynth filter again

I find it very hard to motivate myself to clean up this kind of mess. Porting is easy, messes are hard.

[feisty2]

why not just go Deblock_QED (performance-wise) or Oyster.Deblocking(quality-wise)

[Boulder]

I totally forgot that the original site doesn't have an x64 version, but the Avisynth+ wiki page links to it:

http://www.dropbox.com/s/ui8chlbzopuqs5a/SmoothD2-a3_x64.zip?dl=1

Don't know how messy it is though :)

[Myrsloik]

I totally forgot that the original site doesn't have an x64 version, but the Avisynth+ wiki page links to it:

http://www.dropbox.com/s/ui8chlbzopuqs5a/SmoothD2-a3_x64.zip?dl=1

Don't know how messy it is though :)

That is depressing. Still has inline asm and can only ever be compiled with intel's compiler. And to think this was something produced in modern times...

[Boulder]

Off to check out Oyster then :D

[Boulder]

OK - is there an intermediate choice between Deblock_QED and Oyster? :)

[feisty2]

I could make one if you do want that

[Boulder]

Don't know if you should bother, my needs are only to use a deblocker for those rare TV series that I have only on DVD (Community, MacGyver etc.) Oyster is a bit too much placebo for that usage.

[feisty2]

Guess I'll just give it a shot, I like Community, and it's not that "rare", maybe?

[Boulder]

Thanks, I'd appreciate any effort. If you need any sample material, just let me know.

I feel that TV shows on DVD only are quite rare these days as new series are already in HD when broadcast. The old series need more work though, and they can be quite messy to clean up unless you want to remove all the few details they may still have.

[hydra3333]

I, also, would be economically appreciative. Do you have a paypal donations page of some kind ?

[feisty2]

def fast_deblock(src, radius=6, h=3.2, lowpass="0.0:0.0 0.12:1024.0 1.0:1024.0"):
core = vs.get_core()
MakeDiff = core.std.MakeDiff
MergeDiff = core.std.MergeDiff
MaskedMerge = core.std.MaskedMerge
ShufflePlanes = core.std.ShufflePlanes
colorspace = src.format.color_family
if colorspace != vs.GRAY:
src_color = src
src = ShufflePlanes(src, 0, vs.GRAY)
ref = Oyster.Basic(src, None, radius)
mask = Oyster.helpers.genblockmask(src)
cleansed = Oyster.helpers.nlmeans(ref, 0, 8, 4, h, ref, False)
ref = Oyster.helpers.freq_merge(cleansed, ref, 9, lowpass)
src = Oyster.helpers.freq_merge(cleansed, src, 9, lowpass)
clip = MaskedMerge(src, ref, mask, first_plane=True)
if colorspace != vs.GRAY:
clip = ShufflePlanes([clip, src_color], [0, 1, 2], vs.YUV)
return clip


http://i.imgur.com/buFGWac.png

clip = fast_deblock(clip)

http://i.imgur.com/dYjq5nP.png

basically a low precision lite version of Oyster.Deblocking, lower quality but like a million times faster
note that chroma will not be processed

[Boulder]

Thanks, I'll give it a go as soon as my time allows. Is it possible to use the standard MVTools package in 16-bit depth? I think that 32-bit floating point stuff is also quite placebo for my sources and will only slow things down. I know that you like to be mathematically accurate, there's nothing wrong with that :)

[feisty2]

not possible, there's no MDegrainN in jackoneill's version