LSharpAAF - AntiAliasing Combo

[cretindesalpes]

Originally Posted by markanini

I upped NNEDI3 qual and nszie parameters to 2 and prescrn=false and didnt get much better results

These antialiasing techniques (using edge-directed interpolation) focus on smoothing lines and connecting them. They cannot improve Moiré patterns, they make them even worse because the ghost lines become real.

Originally Posted by thanhtu5013

Your pseudo sounds french, you're french like me, right ?

Saperlipopette, je suis démasqué !

[thanhtu5013]

I have a last question.

Does it possible to make nnedi2 filter more efficient except changing strength values ?

With strh=2 and strv=2, the filter works fine but there has still a bit of aliasing.
It works like a 2x Antialiasing. I'm looking for something like 4x or 8x antialiasing.
Changing strh and strv values doesn't modify anything about that.

@cretindesalpes

Ton filtre marche très bien mais il y'a encore un peu de crénelage.
Changer les valeurs de strh et strv jouent sur le floutage ou le détail mais ne résout pas mon problème.
Peux tu me dire comment je dois modifier mon script pour qu'il applique un Antialiasing 4x ou 8x (si tu joues aux jeux, tu comprends ce que je veux dire) ?

[2Bdecided]

Originally Posted by markanini

Here's an image I'm trying to improve:
http://homepages.ihug.co.nz/~online/...er/blpic-s.jpg

Can't you ask for a bigger copy?

Use TinEye if you can't remember where you got it from.

(sorry if that's a stupid suggestion, but you're never going to get the guitar strings on the neck to look right from that source - the aliasing is too bad).

Cheers,
David.

[markanini]

Youre right about tineye 2Bd, its a great tool. I my case it didnt help but that's the nature of some things...
Do you know any tricks that might help in simmilar situations beside lowpass via spline36resize?

[Shon]

Not downloaded http://www.megaupload.com/?d=LCFXGZL0

[Leinad4Mind]

The link works dude! xD

[mastrboy]

Nice idea, and a great resource.

I kind of miss some documentation of the different AA types, with slowness and quality comparison.

I ran some tests myself on a avs script im working on for a anime dvd rip.

There is alot of others filters also in this script, but here is some fps and cpu data on a 720x480 source. Encode was directly with x264.exe on a i7 860 with disabled HT:

maa      fps: 4,9 | cpu: 30%
ediaa	 fps: 4.6 | cpu: 99%
aaa      fps: 5,7 | cpu: 40%
antialiasing   fps: 5,6 | cpu: 40%
splinaa  fps: 6,0 | cpu: 35%
supaa	 fps: 2,2 | cpu: 60%
naa      fps: 1,9 | cpu: 75%
shaarp   fps: 4,2 | cpu: 55%

(the FPS and CPU load is average over 5min)

[Leinad4Mind]

Thanks mastrboy. Hope you guys like it. ^^

I have update today the script with 1 more GOOD antialiasing: "aaf". The new avsi is attach.

Best Regards

[Leinad4Mind]

New version of it. I've found a bug in Presharpening, I've forget to finish it xD. And I've improve the code a lot ^^.

I've added (bool "stabilize", int "tradius", bool "MT", int "aapel", int "aaov", int "aablk"). tradius can be 1, 2 or 3 for choosing MDegrain1,2 or 3. MT option can be turned ON.
And added some code for motion estimation.
Added NEEDI3 too as AA.

And I have remove some antialiasers from the avs, like maa and daa, 'cause they are in AnimeIVTC... so... But I have added in the Requirements Pack the AnimeIVTC.avsi.

Best Regards and check the 1st page

[supernater]

I get the following error when trying to run LSharpAAF()...

Avisynth script error.
Script Error, there is no function named "MSuper"
(LSharpAAF.v1.2.avsi, line 59)

I copied the required plugins into my plugins folder. Is there something else I'm doing wrong?

[poisondeathray]

Originally Posted by cretindesalpes

NNEDI3 is a CPU hog but gives excellent results. Replace it with NNEDI2 if it takes too long. Here is an example of santiag (3, 3) with the picture you linked on a previous message.

Is there a way to write the santiag function so there isn't a frame shift? (like the way nnedi2_rpow2() has the cshift arugment?)

[Gavino]

Originally Posted by poisondeathray

Is there a way to write the santiag function so there isn't a frame shift? (like the way nnedi2_rpow2() has the cshift arugment?)

As far as I can see, this is already built-in, via the line
Spline36Resize (w, h, 0, -0.5, w, h * 2)
in function santiag_dir.

[poisondeathray]

Originally Posted by Gavino

As far as I can see, this is already built-in, via the line
Spline36Resize (w, h, 0, -0.5, w, h * 2)
in function santiag_dir.

Weird, did you try it out on any sample?

I double checked that I copied the function correctly, and used different sources, they all seem to exhibit the shift (like the old NNEDI before the cshift was implemented)

[Gavino]

Originally Posted by poisondeathray

Weird, did you try it out on any sample?

No, I didn't - I just accepted it at face value.
Looking at it again, I think it should be
Spline36Resize (w, h, 0, +0.5, w, h * 2)

Also it looks to me that strh and strv are the wrong way round - strh affects vertical and strv horizontal.

[poisondeathray]

Yes the "+" fixed it , thanks Gavino . I haven't tested the strh or strv

[poisondeathray]

To cretindesalpes or Gavino or anyone else:

Regarding the santiag() function I have some observations & questions:

When it's a YV12 / YUY2 source, the Spline36Resize (w, h, 0, +0.5, w, h * 2) seems to fix the shift

When it's a RGB24 source, the Spline36Resize (w, h, 0, -0.5, w, h * 2) seems to fix the shift

1) Why would this be?

2) Is it possible to modify the function to automatically adjust for this (maybe if/then statements or conditional ) ? My programming skills are NIL

[elguaxo]

Originally Posted by Gavino

Also it looks to me that strh and strv are the wrong way round - strh affects vertical and strv horizontal.

Yes, indeed.

Source:


santiag(strh=2,strv=10):


santiag(strh=10,strv=2):

[Gavino]

Originally Posted by poisondeathray

Regarding the santiag() function I have some observations & questions:
When it's a YV12 / YUY2 source, the Spline36Resize (w, h, 0, +0.5, w, h * 2) seems to fix the shift
When it's a RGB24 source, the Spline36Resize (w, h, 0, -0.5, w, h * 2) seems to fix the shift

1) Why would this be?
2) Is it possible to modify the function to automatically adjust for this (maybe if/then statements or conditional ) ?

The usual reason for such differences is that Avisynth processes RGB 'upside-down', so the shift is in the opposite direction. However, here I would expect it to 'come out in the wash', since the direction of the shift is a consequence of the use of field=0 in the call to nnedi3(dh=true). Perhaps it is actually a bug in nnedi3's RGB handling. (I didn't even realise till now that nnedi3 accepts RGB, its nnedi predecessors didn't.)

Whether the fault lies with santiag or with nnedi3, a fix/workaround would be
shift = IsRGB() ? -0.5 : 0.5
Spline36Resize (w, h, 0, shift, w, h * 2)

EDIT: Tests suggest this is a fault in nnedi3 - see here.

[poisondeathray]

Thanks Gavino

The cshift parameter for nnedi2_rpow2/nnedi3_rpow2 isn't affected by YV12 /YUY2 vs. RGB ; is the same "workaround" being used there ?

[Gavino]

Originally Posted by poisondeathray

The cshift parameter for nnedi2_rpow2/nnedi3_rpow2 isn't affected by YV12 /YUY2 vs. RGB ; is the same "workaround" being used there ?

I guess that's really a question for tritical - I can only speculate.

What I do see is that if nnedi2/3_rpow2 is called without the cshift parameter, it produces different results for YUY2 and RGB.
That's not necessarily wrong - since that function has no 'field' parameter to say where the original pixels should end up, the function is free to do what it likes - but it seems anomalous.
When cshift is specified, I guess the implementatation takes that difference into account, so the final result is correct.