MCTemporalDenoise [v1.4.20 - Update 2010/07/02]

[Terka]

Hi LaTo,
have you tried the pre-search routine for better motion vectors search introduced by Didee in TGMC?

[Didée]

What for? MCTD already offers "normal" pre-filtering for motionsearch, and that is fully sufficient. The routine in TGMC is custom-tailored to remove the typical on-off-on-off resp. up-down-up-down of spatially-interpolated bobbing. For arbitrary non-interlaced input, there is not any benefit to use that particular method. Chances are that it would come out less good than FFT3D-prefiltering.

For very tough cases, sometimes it is benefitial to use some additional stabilization after FFT3D, since FFT3D might leave back some "nervous" areas that still can mislead the motion engine (see the "HQDN3D" addendum in my predecessor-to-MCTemporalDenoise script). (In MCTD, this is partly achieved via debanding - it's not temporal stabilisation, but helps to dimish the sideffects that *floating* banding imposes on the motionsearch).
But that's big guns for big enemies, and not related to what TGMC is doing.

[Terka]

ok. to be clear for me.
(all following is regarding progressive filtering)
1.temporalsoften routine from TGMC part is worse than ttf3d (or fft3d +HQDN3D)
2.how do you check the quality of motion search? (mean for practical testing-which prefilter to use etc.)

[Didée]

1.temporalsoften routine from TGMC part is worse than ttf3d (or fft3d +HQDN3D)
No, that's a false statement. If it would be "worse" and FFT3D would be "better", then tell me why I didn't use FFT3D within TGMC?

It's not worse, it's just different. For what is to be done in TGMC, the temp-gauss temporalsoften is method-of-choice. For general noise removal prior to motionsearch, FFT3D is the (most versatile) method of choice.

A hammer for the nails, a screwdriver for the screws. Each for it's own purpose.

( I know you said "(all following is regarding progressive filtering)" - but even in TGMC, all filtering is in fact done on progressive frames...)


2.how do you check the quality of motion search?
Eyeballing + experience. Very rarely, vector visualisation.

[Terka]

my problem is i cant recognize the nails/screws without putting the screwdriver on it.

[mozzle]

Are the older versions available somewhere? I'd like to test versions from v1.3.0.0 on to see whether my odd x264 crashes are due to MCTD. It is pretty much the only thing that I have changed recently and debugging x264 pointed the error to be related to Avisynth.


Hey Boulder. Did you figure out if it was MCTD that was crashing x264. I've been having the same crashing issue and had to revert back to x264 v1400.

[Boulder]

No, at least v1.3.0.9 didn't help and that's where the bigger changes were not yet introduced. I'm currently running a test to see whether a different Avisynth version helps (I used to have 2.6.0-MT but now I'm trying 2.5.8-MT). If that doesn't help, I'll try non-ICL builds of x264.

[~SimpleX~]

First of all, thanks for awesome script, LaTo. But I got some problems when tried to encode my video.

That's the problem: http://www.mediafire.com/imageview.php?quickkey=4jzvwmmmy2j

And this's the script I used:


setMTMode(2,0)
SetMemoryMax(256)

src=DGDecode_MPEG2Source("C:\Encoding\Cafe Junkie\VTS_01_1.d2v", info=3, cpu=0)

src=src.ColorMatrix(hints=true, threads=0)

progressive=src.AnimeIVTC(mode=1, aa=1, ifade=true)

ECMode="DFTTest().dehalo_alpha_MT(darkstr=0.7,brightstr=1.7)."+
\ "awarpsharp2(depth=10, blur=2, chroma=2).FastLineDarkenMOD(thinning=12,strength=80, luma_cap=170)"

denoised=progressive.MCTemporalDenoise(\
settings="low", radius=2, GPU=false, fixband=true, sigma=3, chroma=true ,\
edgeclean=true, ECmode=ECMode, ECrad=3, ECthr=16 ,\
enhance=true ,\
protect=true ,\
sharp=true, strength=140 ,\
post=0 ,\
deblock=true )



final=denoised.ediaa().Tweak(sat=1.1).Crop(0,0,-2,0).Spline36Resize(720, 540)

final
__film = last
__t0 = __film.trim(0, 36751)
__t0


What am I doing wrong?

[Boulder]

No, at least v1.3.0.9 didn't help and that's where the bigger changes were not yet introduced. I'm currently running a test to see whether a different Avisynth version helps (I used to have 2.6.0-MT but now I'm trying 2.5.8-MT). If that doesn't help, I'll try non-ICL builds of x264.

It seems that Avisynth 2.6.0-MT causes the problem, no crashing with 2.5.8. I'll have to ask SEt to see if he can find anything wrong there.

[ninja_racoon]

there's a problem when I use MT and MCT+global settings together

MT("MCTemporalDenoise(settings="low")",2,0)

how can I get this to work?

[Gavino]

MT("MCTemporalDenoise(settings="low")",2,0)

how can I get this to work?
Use triple quotes to allow nested string literals:
MT("""MCTemporalDenoise(settings="low")""",2,0)

[Poutnik]

Use triple quotes to allow nested string literals:
MT("""MCTemporalDenoise(settings="low")""",2,0)

The question is, if MT() in conjunction of MC scripts is a good idea.
Most people think it is not, because MC need to cover all spatial scope.

[nautilus7]

I get an error when using settings="very low" or settings= "low". It seems that mode doesn't have the correct value in line 1060:

GFc = (enhance==true) ? TTc.GradFun2DBmod(thr=GFthr,thrC=chroma?GFthr:1.0,mode=0,str=AGstr,strC=0.0,temp=temp,adapt=bias,mask=false,show=false)

http://img511.imageshack.us/img511/5832/errorhm.png

[Blue_MiSfit]

Meh.. I use MT all the time with MDegrain2. It's not optimal, but I've yet to see hideous artifacts.

SetMTMode is probably preferable, but it's caused more problems in my case.

MCTD is fantastic, especially in a pure 64 bit environment with the latest 2.5.8 builds!!!

~MiSfit

[LaTo]

I get an error when using settings="very low" or settings= "low". It seems that mode doesn't have the correct value in line 1060:

GFc = (enhance==true) ? TTc.GradFun2DBmod(thr=GFthr,thrC=chroma?GFthr:1.0,mode=0,str=AGstr,strC=0.0,temp=temp,adapt=bias,mask=false,show=false)

I am getting the following error in MeGUI with 1.4.08:

AviSynth script error:

'mode' have not a correct value! [1,2,3]

(GradFun2dBmod.v.1.5.avsi, line 175)
(MCTemporalDenoise.v.1.4.08.avsi, line 1061)
(MCTemporalDenoise.v.1.4.08.avsi, line 1079)
(C:\PATH TO AVS SCRIPT)

Any help would be appreciated. :thanks:

Re-download GradFun2DBmod v1.5! (here (http://forum.doom9.org/showthread.php?p=1358005#post1358005))

Answer was in the topic, page 15 :D

[nautilus7]

Sorry, my bad.

[Poutnik]

....
MCTD is fantastic, especially in a pure 64 bit environment with the latest 2.5.8 builds!!!

I feel little simple-minded but how is it possible to use MCTD in pure 64bit enviroment ?
I know there is 64bit Avisynth, but are there all needed Avisynth plugins ported to 64bit yet?
It would be great news for me.

[levi]

not to get off-topic, but you can find the 64bit avisynth + 64 bit plugins here (http://forum.doom9.org/showthread.php?t=152800)

[~Revolution~]

I'm not sure if its just me but MDegrain2 with MT always causes a transparent type, very light, grainy horizontal line straight in the middle of the video.

[lych_necross]

How much faster would MCTD perform using a 64bit build of avisynth and 64bit pluigns versus all 32bit avisynth software?

[osgZach]

I'm not sure if its just me but MDegrain2 with MT always causes a transparent type, very light, grainy horizontal line straight in the middle of the video.

Are you sure its not an MT related artifact from the split frame ?

Try using the overlap parameter if you haven't yet.

[Poutnik]

not to get off-topic, but you can find the 64bit avisynth + 64 bit plugins here (http://forum.doom9.org/showthread.php?t=152800)

Is there x64 DctFilter, required by Deblock QED from MCTD ?
I cannot find it either in x64 AVS 2.5.8 MT thread, either on Squid80 site.....

[Didée]

AFAIK, no .... DCTFilter hasn't been ported yet.

But if someone *would* port it, that would be a good occasion to eliminate the stripe-bug that DCTFilter has had ever since. ;)

[LaTo]

Is there x64 DctFilter, required by Deblock QED from MCTD ?
I cannot find it either in x64 AVS 2.5.8 MT thread, either on Squid80 site.....

If you don't use deblocking in MCTD then you don't need DctFilter ;)

[Poutnik]

If you don't use deblocking in MCTD then you don't need DctFilter ;)

I know :) but even with Deblock=false my Vdub64 is crashing after few frames, and avs2avi64 after some dozens of frames.

later I will try by cut away method to detect what plugin causes it.
AVS64 itself seems to be fine.

It is a pitty, because MCTD32 is nice and stable, but little speeding up would be handful.

[greatmagician]

thanx brother .... I have ? here
LaTo this working with the mod 16
in this case i ca'nt edit the size for ex: I got with mod16 (1280 X 528 ) I got DAR 2.42.1 > but i want to fix it it must be in DAR 2.35.1 ( 1280 X 534 )
but the problem i can't because I will get the error :
This is Not a MOD4 clop! >> in this case what I will do ??!!

thanx alot LaTo ...

[LaTo]

thanx brother .... I have ? here
LaTo this working with the mod 16
in this case i ca'nt edit the size for ex: I got with mod16 (1280 X 528 ) I got DAR 2.42.1 > but i want to fix it it must be in DAR 2.35.1 ( 1280 X 534 )
but the problem i can't because I will get the error :
This is Not a MOD4 clop! >> in this case what I will do ??!!

thanx alot LaTo ...
Width and height must be divisible by 4.

or workaround:
Addborders(0,2,0,0)
MCTemporalDenoise(...)
Crop(0,2,0,0)

[greatmagician]

Width and height must be divisible by 4.

or workaround:
Addborders(0,2,0,0)
MCTemporalDenoise(...)
Crop(0,2,0,0)

thnx LaTo but :confused: but the DAR same ( 2.42:1 )

this my script :
crop( 0, 142, -2, -142)
Spline36Resize(1280,528)
Grainfactory3(1,1,1)
GradFunkMirror()
MCTemporalDenoise(settings="high")
how i will fix it i want to edit my Resize but i can't

LaTo can u mke update for (MCTemporalDenoise)
we want to make the control for The script not MCTemporalDenoise ... :) thnx

[LaTo]

crop( 0, 142, -2, -142)
Spline36Resize(1280,528)
Grainfactory3(1,1,1)
GradFunkMirror()
MCTemporalDenoise(settings="high")
Adding grain+dither before MCTD is useless...

Try:
crop( 0, 142, -2, -142)
Spline36Resize(1280,534)

Addborders(0,0,0,2)
MCTemporalDenoise(settings="high")
crop(0,0,0,-2)

Grainfactory3(1,1,1)
GradFunkMirror()

[Gannjunior]

Hi guys,

what do you use to fast your encoding with MCTtemporalDenoise?

This is my scrypt, where i'm trying to use MT :


SetWorkingDir("d:\avs_plugins\AviSynthPlugins\")
LoadPlugin("MT.dll")
Loadplugin("d:\avs_plugins\warpsharp.dll")

LoadPlugin("SangNom.dll")
LoadPlugin("average.dll")
LoadPlugin("Degrainmedian.dll")
Loadplugin("removegrain.dll")
LoadPlugin("Repair.dll")
LoadPlugin("gradfun2db.dll")

LoadPlugin("DGAVCDecode.dll")
LoadPlugin("DGDecode.dll")

loadplugin("fft3dfilter.dll")
loadplugin("hqdn3d.dll")
LoadPlugin("convolution3d.dll")
LoadPlugin("undot.dll")
LoadPlugin("mvtools2.dll")
LoadPlugin("mt_masktools-26.dll")
LoadPlugin("TTempSmooth.dll")

import("d:\avs_plugins\denoisers\MCTemporalDenoise.v1.4.08.avsi")
import("d:\avs_plugins\denoisers\GradFun2DBmod.v1.5.avsi")
import("LSFmod.v1.9.avsi")
LoadPlugin("GradFun2DB")

LoadPlugin("D:\Program Files (x86)\StaxRip\Applications\AviSynth plugins\DGAVCDecode.dll")

AVCSource("L:\test_5d2\MVI_0135 temp files\MVI_0135.dga")
MCTemporalDenoise(settings="high")

I tried to use SetMTmode() in different way but with no success.

How could I set MT in the correct way to use all my i-920 cores in the script above?
Or, if you could suggest me alternative paths to use my i7, i would be happy!

Thanks in advance

---------------

P.S. "settings high" is the extreme setting of MCTtemporalDenoise? because i want to test it on my video shooting at 12800 iso.

ciao! :)

[Nephilis]

Try MT("""MCTemporalDenoise(settings="high")""",threads=4) instead of SetMTMode()..

[Gannjunior]

thanks for your suggestion.
it starts and it gets 20% of cpu more or less...the encoding anyway takes me jurassic time.
if u have any other suggestion i'm happy to test it! :-)
ciao!

[BigDid]

Hi there,

it starts and it gets 20% of cpu more or less...the encoding anyway takes me jurassic time.
It works so it seems an improvement. What if you do not use MT, just:
MCTemporalDenoise(settings="high")
will your encoding time be paleolithic (even slower as it should be) ? ;)
If not slower something is wrong in your config and/or settings
Be aware that MCTD is a slow (to very slow) filtering process!


if u have any other suggestion i'm happy to test it! :-)
ciao!
What avisynth rev and MT dll are you using?
If not 2.58 MT from SET (MT-avisynth+ MT dll) or 2.6 MT-avisynth from SET you may not have stability or performance accordingly...

For an i7, you could try to change the number of threads from 4 to 6 to 8 or even more and see if any speed improvement.

You did not state if you are encoding to xvid, x264 or other codec and what GUI you are using for that purpose; vfw xvid/divx and gui are known to be suboptimal compared to x264 and non vfw GUI for multicores ...

Did

[Gannjunior]

i'm using avisynth dll and MT dll updated to 2010 jan.

tried to set 8 core but no speed improvement

since my "raw" source is 45mbit avchd from 5d mark II, i usually recompress to h264.

Since the direct conversion to h264 of the script above in megui is more tha paleolithic, i'm encoding (now) in lagarith loseless by vdub. (passed 5h30' and i encoded 1100 frames !!!).

so you suggest to set MCTtempDen to "medium" settings? do i get an heavy denoise the same?

because from some preview i've seen that setting "high" seems to be amazing. It really drops alla the noise produced by my camera at 12800 iso but i can see a really amount of detail that camera captured.

thank you very much, ciao!!

[BigDid]

Hi,

This:
i'm using avisynth dll and MT dll updated to 2010 jan.
is not SET build. Use it at your own risk :rolleyes:

Rather, see:
http://forum.doom9.org/showthread.php?t=148117


tried to set 8 core but no speed improvement

Ok.
I also asked you to test with and without MT to be sure your avisynth+MT configuration is correct and you have a REAL speed gain with MT.
From a quick and dirty test encode with my Q9400 @3200 (1pass-Cq3, Xvid-encode, SD):

MCTemporalDenoise(settings="high")............ is +/- 5.5 fps
MT("""MCTemporalDenoise(settings="high")""") is +/- 9.5fps


so you suggest to set MCTtempDen to "medium" settings? do i get an heavy denoise the same?

Where have you seen me suggesting that? please quote :mad:
Seriously, this is the MCTD thread, I answered on your MT problem only. For MCTD, review posts on differences between settings and tuning by LATO, very interesting.

Did

[Gannjunior]

Sorry maybe i read in the not correct way what you wrote before!

FRom what you are sayng there is something wrong in my config, beacuse with my i7 i have to go at least at your speed...And i'm really far from 5 fps...since i don't reach 1fps !!!

which MT and avisynth .dll do you use?

Which problem do you think i could have in my config?


many thanks in advance and have a nice weekend!!
ciao!

[BigDid]

Hi,

Sorry maybe i read in the not correct way what you wrote before!
Ok, let's say we both need to read or write the correct way. I have edited my previous post accordingly.


which MT and avisynth .dll do you use?
Link was already given for Mt-avisynth 2.58 and MT.dll from SET; once again:
http://forum.doom9.org/showthread.php?t=148117


FRom what you are sayng there is something wrong in my config, beacuse with my i7 i have to go at least at your speed...And i'm really far from 5 fps...since i don't reach 1fps !!!
Yes it seems so even if my test was SD-Xvid encode.


Which problem do you think i could have in my config?
Bad installation or use?
See basics:
http://avisynth.org/mediawiki/MT_support_page
and/or
http://avisynth.org/mediawiki/MT

This is still MCTD thread and if you need more tips for MT, you'd better continue in the SET MT thread or create one (I will not post in Jeremy's thread as I don't use it).

Did

[Assassinator]

The AA portion of MCTD seems to have broken between 1.3.09 and 1.4.08... Or am I just doing something really wrong?


New behaviour ==> http://zingaburga.com/assassinator/ImageStash/Encoding/MCTD1408AA.png

Old behaviour ==> http://zingaburga.com/assassinator/ImageStash/Encoding/MCTD1309AA.png



MCTemporalDenoise(aa=true,useEEDI2=false,maxd=48,aathr=16,stabilize=true)



Oh yeah, these images are produced for the sole purpose of explaining what I'm talking about in this post, so please don't tell me I should be using aa/maa/daa/whatever instead, or I shouldn't be using MCTD/denoise/whatever, since I'm not even going to be encoding this stuff.

[LaTo]

The AA portion of MCTD seems to have broken between 1.3.09 and 1.4.08... Or am I just doing something really wrong?


New behaviour ==> http://zingaburga.com/assassinator/ImageStash/Encoding/MCTD1408AA.png

Old behaviour ==> http://zingaburga.com/assassinator/ImageStash/Encoding/MCTD1309AA.png



MCTemporalDenoise(aa=true,useEEDI2=false,maxd=48,aathr=16,stabilize=true)



Oh yeah, these images are produced for the sole purpose of explaining what I'm talking about in this post, so please don't attempt to tell me I should be using aa/maa/daa/whatever instead, or I shouldn't be using MCTD/denoise/whatever, since I'm not even going to be encoding this stuff.

Maybe broken, I made a lot of changes between v1.3 and v1.4...
I will look at this next month, study now :D

[Assassinator]

Maybe broken, I made a lot of changes between v1.3 and v1.4...
I will look at this next month, study now :D

Thanks :)

No rush, your studies are more important.

[newface]

While i am adding to script MCTemporalDenoise(settings="high") getting error ther is no function named "Msuper" plz rectifie this error image attached below

http://i42.tinypic.com/htchf5.jpg

[um3k]

Sounds like you need MVTools 2: http://avisynth.org.ru/mvtools/mvtools2.html

[newface]

Sounds like you need MVTools 2: http://avisynth.org.ru/mvtools/mvtools2.html

Please give me more detiles

[MatLz]

Please give me more detilesMSuper is one (surely the most important) of the functions in the MVTools2.dll plugin

[jordisound]

Try MT("""MCTemporalDenoise(settings="high")""",threads=4) instead of SetMTMode()..

well, it works. With this sintaxis the speed increase, the 4 cores are working fine.

I am testing gpu=true vs. gpu=false (fft3dgpu vs fft3dfilter). fft3 gpu is 7% faster. analyzing the final video, the compressibility is 5% better with fft3dfilter. it says that the default values for fft3d and fft3gpu are different? or fft3d is more efficient than fft3dgpu?

[Poutnik]

well, it works. With this sintaxis the speed increase, the 4 cores are working fine.........

Interesting, I went the opposite direction MT() -> SetMTMode, realizing the spatial dividing by MT() was causing for me artefacts.

MC is said - seems obvious - to have gain from evaluation the whole picture.

[Joachim Buambeki]

I want to use MCTemporalDenoise to denoise my timelapse footage, but it always give me an error:There is no function named "Sangnom"


The script is just:MCTemporalDenoise(settings="medium") for a 720p clip. As far as I understand there should not even be a call for Sangnom. I don't have the Sangnom.dll, since I am using the 64bit package from JoshyD.

Another question I have is:
What settings would you recommend me to denoise 4k footage, based on the documentation I figured larger block sizes should do no harm at 4k:MCTemporalDenoise(settings="medium", protect=true, stabilize=true, enhance=true, bwbh=32, owoh=16, blksize=32, overlap=16)Is this likely to work? I am afraid such a complex script will will not run very stable at 4k.

Best Regards
David

[LoRd_MuldeR]

http://manao4.free.fr/SangNom.zip

[Joachim Buambeki]

Thanks, but I need a 64bit Sangnom (though I wonder why Sangnom is needed for the above script. :confused:
From what I have read other people use MCTemporalDenoise with succes on with 64bit Avisynth.

Best Regards
David

[LoRd_MuldeR]

Thanks, but I need a 64bit Sangnom (though I wonder why Sangnom is needed for the above script. :confused:

Well, if MCTemporalDenoise makes use of the Sangnom plugin, then you'll need that plugin for MCTemporalDenoise to work. It's that easy :p

However I don't know if a 64-Bit build of Sangnom exists...