HAvsFunc

[Reel.Deel]

Hmmm, I guess I must of missed that... I guess now the only Avisynth plugin required is DCTFilter?... Cool beans

[Mystery Keeper]

Also Deblock. But I tried to replace lutspa with those functions, and it spawns threads like crazy. Right now VirtualDub is running 1581 thread and hanging. My own WIP editor dies quickly after spawning ~700.

[Reel.Deel]

Also Deblock.

Of course http://www.smileyvault.com/albums/basic/smileyvault-slaphead.gif (http://www.smileyvault.com/) (how did I forget that?)

[HolyWu]

Also Deblock. But I tried to replace lutspa with those functions, and it spawns threads like crazy. Right now VirtualDub is running 1581 thread and hanging. My own WIP editor dies quickly after spawning ~700.
Try havsfunc-r7.7z (https://www.mediafire.com/?ynk5rizd26voxf7) and see if it works for you. I didn't give it a heavy test though.


Added the following function: GrainStabilizeMC.
Changed the way to generate block clip instead of using mt_lutspa, which is used in Deblock_QED.
HQDering mod updated to v1.8.
Added sigmoid curves to GammaToLinear/LinearToGamma.
LSFmod's Smode 2 is deleted. Now Smode=1 is Range sharpening, Smode=2 is Nonlinear sharpening (corrected version).
Optimized the way to build LUT of Smode=2 in LSFmod. In previous version std.Expr is used for >8-bit clip and the speed is very slow. Now std.Lut is used for 8~16-bit clip.
Replaced MakeDiff/AddDiff with internal function std.MakeDiff/std.MergeDiff instead of using std.Expr.

[Mystery Keeper]

Works, but the same way as with mt_lutspa. If I seek all the way to the last frame - I crash. But now I did get an error. "Seek pattern broke d2vsource." -_-
But it does work, so thank you again.

[HolyWu]

Updated r8.

Changed Deblock_QED to use native plugins. That's it.

[Mystery Keeper]

Thank you!

[YamashitaRen]

Thank you!
As a linux user, I greatly appreciate these steps toward a fully native vapoursynth :)

[HolyWu]

Updated r9.

Added GrainFactory3 function. The default of g?str is adjusted to a saner value as the default in the original is way too overkill. Besides, g?tex parameter is removed.

[HolyWu]

Updated r10.

Changed QTGMC and some other functions to use native filters of MVTools.

[Mystery Keeper]

Cool! Is it completely native now I thought there were some functions unported yet.

[HolyWu]

Of course not. Only filters that listed at https://github.com/dubhater/vapoursynth-mvtools/blob/v1.1/readme.rst get changed.

[Mystery Keeper]

Can you please give a link for r9 too? So far VS MVTools is buggy, and QTGMC crashes even though I'm not using DCT.

[HolyWu]

No problem. First post is updated. It may be helpful to let jackoneill know what preset or arguments you specified to make it crash. I just let it run on a 1920x1080 clip about 10000 frames with all default values and don't encounter crash. The average speed is roughly at 5.0 fps.

[lansing]

I tried the new version with native mvtools added, on my 720x480 video, qtgmc(preset="medium"),

vapoursynth 32bit: 94fps
vapoursynth 64bit: 94fps

avisynth mt mvtools: 125fps


Updated the result for Vapoursynth 64 bit version, kind of disappointed. I was expecting it to be much faster.

[HolyWu]

Updated r11.

The mvtools-related filters used in QTGMC and ivtc_txt60mc are all native now.

Because QTGMC's SVThin and Lossless modes require rgvs.VerticalCleaner, I also include RemoveGrainVS.dll in the archive so you don't have to wait for the next release of VapourSynth.

[Myrsloik]

Why do we have to do?

import havsfunc
...
haf = havsfunc.HAvsFunc()
...
clip = haf.QTGMC(clip, Preset='Medium', TFF=True)

Wouldn't it make more sense to remove the HAvsFunc class that effectively does nothing so we can do:

import havsfunc as haf
...
clip = haf.QTGMC(clip, Preset='Medium', TFF=True)

[HolyWu]

You are right. The class was used to let users pass the core instance to the module before VapourSynth R19. Since the core now uses a singleton pattern, the class becomes redundant. I just didn't think of doing the modifications though...

[HolyWu]

Update r12.


The class is ditched so you can reduce one line in your script.
Add function Vinverse2.
Delete function InterFrame.
DeHalo_alpha: Remove the color family restriction of input clip.
HQDeringmod: Remove the color family restriction of input clip. Remove dering kernel 'dfttest' and related parameters. Remove stack16 related parameters.
QTGMC: Remove EdiMode 'nnedi2', 'nnedi', 'eedi2' and 'tdeint'.
Vinverse: Remove the color family restriction of input clip.
GSMC: Use native filters of MVTools. Remove lsb parameter.
SMDegrain: Use native filters of MVTools. Remove stack16 related parameters. Change prefilter=3 to MinBlur(3).
FastLineDarkenMOD: Remove the color family restriction of input clip.
LSFmod: Remove the color family restriction of input clip. Change the type of parameter 'preblur' from string to bool.

[jackoneill]

Is your vinverse different from the one in the VapourSynth source tree?

[Reel.Deel]

Vinverse2 is a modified version of Vinverse used as a helper this dot crawl removal script. (http://forum.doom9.org/showpost.php?p=1584186&postcount=62) tp7 also included a fast implementation in his Vinverse plugin. (https://github.com/tp7/vinverse)


Edit: oops, I completely misunderstood jackoneill's question. I honestly thought he asked about vinverse2. The eggnog must of been a little stronger this year ;)

[feisty2]

edit: okay, it's wrong, just tried and the result is completely different from mdegrain
I thought something from here (http://forum.doom9.org/showthread.php?t=156770&highlight=mdegrain9) could be useful
any luck with mcompensate + templinearappoximate to imitate "mdegrain (lsb=true)" ?

[HolyWu]

Is your vinverse different from the one in the VapourSynth source tree?
I think the only difference is high bitdepth support.

[MonoS]

I've also ported a avs script to vs, if you want you can add it to your collection [so that as much script will be available in only one place].
The original script is here http://recensubshq.forumfree.it/?t=64839203
The code is here: https://github.com/MonoS/VS-MaskDetail
The thread is here: http://forum.doom9.org/showthread.php?t=171546

[lansing]

Hi, I'm getting an error while using deblock_qed.

No attribute with the name dct exists. Did you mistype a plugin namespace

ret = haf.Deblock_QED(ret, quant1=40, quant2=24)

I'm using the latest Vapousynth, mvtools and deblock plugin.

[Are_]

dctfilter (http://forum.doom9.org/showthread.php?t=171039)

Next time you can search for the missing dll here (http://www.vapoursynth.com/doc/pluginlist.html).

[lansing]

thanks, it works now

[Mystery Keeper]

MVTools now support up to 16 bits and other formats. Functions need to be fixed to add that support too.

[HolyWu]

Update r13.


Remove the format restriction of the following functions: QTGMC, GSMC, SMDegrain.
SMDegrain: Use sbr for prefilter instead of MinBlur.

[zerowalker]

SMDegrain acts a bit different compared to the Avisynth version on the same settings.
Is there some Default settings that are different?

SMDegrain(tr=2,thSAD=500,search=3)

That's what i tested and compared with.

[HolyWu]

SMDegrain acts a bit different compared to the Avisynth version on the same settings.
Is there some Default settings that are different?

Nope. The only big difference should be the prefilter setting which has been changed to use sbr for prefiltering. The others should be quite the same.

Test the core mvtools part first on your own to check whether the issue comes from mvtools itself. Here are equivalent settings for AVS and VS respectively.


super = MSuper(rfilter=4)
bv2 = MAnalyse(super, isb=true, delta=2, overlap=4, search=3, truemotion=false)
bv1 = MAnalyse(super, isb=true, delta=1, overlap=4, search=3, truemotion=false)
fv1 = MAnalyse(super, isb=false, delta=1, overlap=4, search=3, truemotion=false)
fv2 = MAnalyse(super, isb=false, delta=2, overlap=4, search=3, truemotion=false)
output = MDegrain2(super, bv1, fv1, bv2, fv2, thSAD=500, thSADC=250)



super = core.mv.Super(clip, rfilter=4)
bv2 = core.mv.Analyse(super, isb=True, delta=2, overlap=4, search=3, truemotion=False)
bv1 = core.mv.Analyse(super, isb=True, delta=1, overlap=4, search=3, truemotion=False)
fv1 = core.mv.Analyse(super, isb=False, delta=1, overlap=4, search=3, truemotion=False)
fv2 = core.mv.Analyse(super, isb=False, delta=2, overlap=4, search=3, truemotion=False)
output = core.mv.Degrain2(clip, super, bv1, fv1, bv2, fv2, thsad=500, thsadc=250)

[zerowalker]

Tested, it looks a bit different.
It's like the image is pushed up 1 line in some places VapourSynth (in a stretched way).

Does yours look identical?

[zerowalker]

So what am i supposed to do?
They looked different, is it supposed to, i mean it doesn't use the prefilter as far as i can tell, if rfilter isn't that of course? (isn't Prefilter when you add an already filtered imaged and use it as some kind of layer or something).

[Reel.Deel]

(isn't Prefilter when you add an already filtered imaged and use it as some kind of layer or something).

A prefiltered clip is usually used in order to try to get better motion analysis; gather motion vectors from the prefiltered clip and then use those vectors on the original clip.

[feisty2]

prefilter is a special filter used only on vector search clip
when you do the motion compensation (mv.compensate/degrain/...), vectors are required to move blocks from predicted frames to the current reference frame
the point is, you don't have to get vectors from the clip you are performing compensation
like, you can get vectors from a spatial denoised version of the source clip so you can reduce the local SAD value
that's pretty much it

[zerowalker]

So how come my results look different between them if i don't use a prefiltered source to do the vector thing?
I am missing something here, shouldn't they be identical?

[jackoneill]

So how come my results look different between them if i don't use a prefiltered source to do the vector thing?
I am missing something here, shouldn't they be identical?

Not 100% identical, no. There are some very small differences in the output of Super, which I haven't bothered to investigate. The small differences in Super propagate to every other filter in MVTools. However, I don't think these particular differences are visible in the output of Degrain, so it must be something else. Please show some screenshots (original image, filtered with Avisynth MVTools, filtered with VapourSynth MVTools). Maybe we should take this to the MVTools thread (http://forum.doom9.org/showthread.php?t=171207&page=6).

[Tarutaru]

Hi, QTGMC won't start and throw this error
$ vspipe Untitled.vpy -
Segmentation fault (core dumped)

and I got this in dmesg
traps: vspipe[17476] general protection ip:7f1b98e4d6cd sp:7f1b8e384c30 error:0 in libgenericfilters.so[7f1b98e48000+10000]

Is there something wrong with the genericfilters or some other problem? Thanks!
(OS - Linuxmint 17.1 x64)

Script -
import vapoursynth as vs
import havsfunc as haf
core = vs.get_core()
core.std.LoadPlugin('/usr/lib/x86_64-linux-gnu/libffms2.so.3')
ret = core.ffms2.Source(source='foo.bar')
ret = haf.QTGMC(ret, Preset='Medium', TFF=True)
ret.set_output()

[Tarutaru]

Solved by using Myrsloik's fork.
Thanks!!

Hi, QTGMC won't start and throw this error

and I got this in dmesg


Is there something wrong with the genericfilters or some other problem? Thanks!
(OS - Linuxmint 17.1 x64)

Script -

[HolyWu]

Update r14.


Add functions SigmoidInverse and SigmoidDirect from Dither package.
Remove the bitdepth restriction of ediaa function.
For functions that use mvtools, now internally apply DitherLumaRebuild to the clip when it's being fed to 'mv.Super' (for the final 'mv.Analyse' only). s0=1 is always used.
QTGMC: Use native filter of FFT3DFilter for noise processing if selected.

[HolyWu]

Update r15.


HQDeringmod: When invoking generic.Sobel, mthr was not properly scaled with the bitdepth of input clip.
QTGMC: When invoking TemporalSoften, thresholds and scenechange were not properly scaled with the bitdepth of input clip.

[Pat357]

Update r15.


HQDeringmod: When invoking generic.Sobel, mthr was not properly scaled with the bitdepth of input clip.
QTGMC: When invoking TemporalSoften, thresholds and scenechange were not properly scaled with the bitdepth of input clip.


Would you consider adding MCBob to HAvsFunc ?
Info & code about MCBob : http://avisynth.nl/index.php/MCBob .

For me the script is really to complicated to rewrite it myself ; otherwise I would have done it already.

Thanks !

[HolyWu]

Would you consider adding MCBob to HAvsFunc ?
Info & code about MCBob : http://avisynth.nl/index.php/MCBob .

For me the script is really to complicated to rewrite it myself ; otherwise I would have done it already.

Thanks !

I don't know, probably no. Isn't MCBob superseded by TempGaussMC which is considered to be more stable?

[HolyWu]

Update r16.


Add functions logoNR and SmoothLevels.
maa: Use native filter of SangNomMod. Expose 'aa' and 'aac' parameters.
SharpAAMCmod: Use native filter of SangNomMod if selected.
ivtc_txt60mc: Fix misusing of DitherLumaRebuild.
SMDegrain: Update to v3.0d.
Cosmetic changes.

[MonoS]

I think there's a little bug into QTGMC.

When using InputType 2-3 at line 1013 got passed to mv.Mask a 16bit clip, but the plugin require a 8bit clip
Solved dithering srchClip like so [i mostly use 16bit QTGMC so i'm fine with this change]
inputTypeBlend = core.fmtc.bitdepth(core.mv.Mask(core.fmtc.bitdepth(srchClip, bits=8), bVec1, kind=1, ml=ProgSADMask), bits=16)

There's another instance in the code of mv.Mask, but wasn't triggered by my script

[HolyWu]

It looks like jackoneill forgot to update mv.Mask to support 16 bits as well. :D

[jeremy33]

Hello,

I try to use LSFMod but I have this error "Erreur de segmentation" (Segfault)
If I try clip = haf.QTGMC(clip, Preset='Medium', TFF=True) I have this error "Exception en point flottant" (Floating point exception)

What's the problem ?

Thank you

[djcj]

What license is HAvsFunc?

[HolyWu]

Hello,

I try to use LSFMod but I have this error "Erreur de segmentation" (Segfault)
If I try clip = haf.QTGMC(clip, Preset='Medium', TFF=True) I have this error "Exception en point flottant" (Floating point exception)

What's the problem ?

Thank you

Sorry, I can't reproduce it on both 32 and 64 bits. What version of Python and VapourSynth did you install? Did you use vspipe to test your script? Normally you should get a traceback pointing out the code at which line caused the exception.

What license is HAvsFunc?

No license, just like most of the AVS scripts.

[jeremy33]

Sorry, I can't reproduce it on both 32 and 64 bits. What version of Python and VapourSynth did you install? Did you use vspipe to test your script? Normally you should get a traceback pointing out the code at which line caused the exception.
I'm on Netrunner 14 x64 (Ubuntu 14.04 derivative) and I use Python 3.4.0 and VapourSynth 27.

I use MPV, I also tried with Vspipe, and I only have this error with LSFMod : Segfault