Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Capturing and Editing Video > Avisynth Usage

Reply
 
Thread Tools Search this Thread Display Modes
Old 29th October 2009, 21:43   #1  |  Link
bizz & buzz
Banned
 
Join Date: Jun 2008
Posts: 94
TempGaussMC "insane" settings?

Hi all,
According to this post, the Default settings of TempGaussMC are a medium effort. So Iv'e been wondering what are the extreme settings of TempGaussMC, visual quality wise? to be honest, reading to docs and figuring this out for myself is still a bit confusing...

Last edited by bizz & buzz; 30th October 2009 at 04:36.
bizz & buzz is offline   Reply With Quote
Old 30th October 2009, 09:49   #2  |  Link
Terka
Registered User
 
Join Date: Jan 2005
Location: cz
Posts: 704
1. ON which source, what artefact do you want to remove, or what you want to achieve?

2. Have some experience on DV sources.
tr0=2
tr1=2
tr2=3
(or more-you can change the script, so it uses mdegrain5
EdiMode="nnedi2" - or you can change the script to use
TdeintTMM_Nnedi2:
Quote:
function TdeintTMM_Nnedi2(clip c)
{
GetParity(c) ? AssumeTFF(c) : AssumeBFF(c)
Tdeint(c,mode=1,emask=TMM(mode=1),edeint=nnedi2(c,field=-2))
}
search=3 exhaustive search
DCT=0 <- try yourselves 1 or more, but speed is getting looow (depending also on blocksize)
the blocksize change wont help
instead of inner sharpen routine use SuperSlowSharpen
tryied also mrecalculate using blocksize=32 and blocksize=4

IMHO the red will help more, the orange will help less, other are discutable.
Myself spend a lot of time playing with it. But im rookie.
You can also try modify MCbob:
return (last) -> return (repaired) (and also put in it nnedi2 instead nnedi)

IMHO conclusion: insane settings and tweaks are possible, but you got 2% in quality for 5-10x slower time.
So until someone will get an idea, how get the 2% for 2x slower time, or mvtools will significantly improve or new filters appear...use the default, or red, or play with the script yourselves
Terka is offline   Reply With Quote
Old 30th October 2009, 20:11   #3  |  Link
bizz & buzz
Banned
 
Join Date: Jun 2008
Posts: 94
I'm trying to retain as much details as possible (i.e no denoising higher then defaults) and still get a stable and accurate result as much as possible. Speed is not an issue - I'm willing to leave the machine working for days.

Here is a little report after testing some of your your recommendations:
- tr2 - setting it to 3 is probably more precise then the default (1), but it also denoise more, so no go.
- EdiMode settings are source dependent (here). on my specific source "nnedi2" was better then EEDI2 and Yadif.
- Search=3 yield very small to no improvement.
- Changing DCT didn't make much difference.
- Blocksize - reducing it to 8 actually created some minor distortion on horizontal straight lines in some frames. Setting it back to the default 16 resolved the issue.
Haven't got time to try SuperSlowSharpen and the modified MCbob

So to my eyes, apart from EdiMode (and TGMC's Sharpness), changing the other parameters indeed made only a small difference.

Thanks for your input Terka, and if there's some more suggestions I'll be happy to try them.
bizz & buzz is offline   Reply With Quote
Old 30th October 2009, 20:53   #4  |  Link
Didée
Registered User
 
Join Date: Apr 2002
Location: Germany
Posts: 5,389
You might want to give a try on that noise-bypass technique with TGMC.
__________________
- We´re at the beginning of the end of mankind´s childhood -

My little flickr gallery. (Yes indeed, I do have hobbies other than digital video!)
Didée is offline   Reply With Quote
Old 31st October 2009, 10:08   #5  |  Link
Terka
Registered User
 
Join Date: Jan 2005
Location: cz
Posts: 704
want ask:
1. what is your source?
2. what is the output framerate?
Try the above 'noise-bypass technique' together with tr2=3 (and maybee Blocksize=32 will calm also a bit more)

Last edited by Terka; 31st October 2009 at 10:11.
Terka is offline   Reply With Quote
Old 31st October 2009, 10:15   #6  |  Link
LaTo
LaTo INV.
 
LaTo's Avatar
 
Join Date: Jun 2007
Location: France
Posts: 701
Just a personal taste: Sbb=0 & SVthin=0
LaTo is offline   Reply With Quote
Old 31st October 2009, 14:07   #7  |  Link
bizz & buzz
Banned
 
Join Date: Jun 2008
Posts: 94
The noise-bypass technique did really really well!
also, using tr2=3 with the above technique restored the details that otherwise would have been lost due to tr2=3 - so now tr2=3 will be used.
Changing Sbb didn't change thing significantly, but reducing SVthin did improve the picture a bit, so it will be set to 0.5.

BTW, the source is an NTSC DVD of old NBA games, so output frame rate will be 29.97 (I'll use SelectEven).

some questions:
Is the noise-bypass technique can be useful with other filters that tend to remove details in order to do their job?

Am I right to assume that now there is no need to change block size to 32 (and risk loosing accuracy), because the fine details have already been restored?
bizz & buzz is offline   Reply With Quote
Old 31st October 2009, 20:21   #8  |  Link
Terka
Registered User
 
Join Date: Jan 2005
Location: cz
Posts: 704
the blocksize 32 will make static areas more calm, moving areas will be less precise, but you wont realise this much.
Try yourselves.
If the output is 29.97,
you can try:
Quote:
avisource, separatefields
>nnedi2 alone
or
>TGMC (with loower thSAD, or other denoiser)
so dont use selecteven on the end.
but if its fast, and NBA is damned fast, imho ill left the double framerate.

Last edited by Terka; 31st October 2009 at 20:29.
Terka is offline   Reply With Quote
Old 1st November 2009, 00:12   #9  |  Link
bizz & buzz
Banned
 
Join Date: Jun 2008
Posts: 94
I'm getting an error when setting blocksize to 32:
MVAnalyse: block's size must be 4x4 8x4....32x16.
?
(Couldn't find anything useful with search or Google).

Using NNEDI2 alone was a lot faster then TGMC but as expected, TGMC had better quality.

I too think that sport vids can benefit from the doubled frame rate, so I'll do two encodes and see if the visual improvement will be worth the extra bits.
bizz & buzz is offline   Reply With Quote
Old 1st November 2009, 06:22   #10  |  Link
Terranigma
*Space Reserved*
 
Terranigma's Avatar
 
Join Date: May 2006
Posts: 953
Quote:
Originally Posted by bizz & buzz View Post
I'm getting an error when setting blocksize to 32:
MVAnalyse: block's size must be 4x4 8x4....32x16.
?
(Couldn't find anything useful with search or Google).
Update to the latest version of mvtools (if you haven't by already), and use the updated mvtools2 version i've attached to my post.
syntax to use is practically the same, but with the letter 'u' added at the end.
Attached Files
File Type: zip TempGaussMVT2.zip (5.5 KB, 318 views)
Terranigma is offline   Reply With Quote
Old 1st November 2009, 22:03   #11  |  Link
bizz & buzz
Banned
 
Join Date: Jun 2008
Posts: 94
Working perfectly - thanks Terranigma
bizz & buzz is offline   Reply With Quote
Old 2nd November 2009, 11:46   #12  |  Link
bizz & buzz
Banned
 
Join Date: Jun 2008
Posts: 94
@Terranigma:

There's a small smoothing effect when using TempGaussMC_beta1u and not the regular TempGaussMC_beta1.

Since the difference is only by using MVtools2 and not MVtools 1.XX, the cause for the effect is either a change in MVtools' internal functions, or a change in it's defaults. Looking at the defaults, I did find some changes (Search, Rfilter) but I'm not sure they are related.

is there any way to eliminate this smoothing effect?
bizz & buzz is offline   Reply With Quote
Old 2nd November 2009, 15:47   #13  |  Link
Terranigma
*Space Reserved*
 
Terranigma's Avatar
 
Join Date: May 2006
Posts: 953
Quote:
Originally Posted by bizz & buzz View Post
@Terranigma:

There's a small smoothing effect when using TempGaussMC_beta1u and not the regular TempGaussMC_beta1.

Since the difference is only by using MVtools2 and not MVtools 1.XX, the cause for the effect is either a change in MVtools' internal functions, or a change in it's defaults. Looking at the defaults, I did find some changes (Search, Rfilter) but I'm not sure they are related.

is there any way to eliminate this smoothing effect?
Perhaps it is rfilter (since it uses a different smoothing method as default). I was going to mod it to use the old method [rfilter 0] and add the rfilter parameter as an configurable option, but when I was comparing the mvtools2 version with the default smoother rfilter [rfilter 2] against the old mvtools method [rfilter 0], I haven't noticed much difference (but maybe you have better sight than i do =P).

What you could do however is open the script with a text-editor such as notepad or wordpad and hit ctrl+f to bring up the search menu and do a search for msuper (3 total) and add rfilter and set it to mode 0 for all 3 (per the old method in mvtools 1)--see if that helps.
Terranigma is offline   Reply With Quote
Old 2nd November 2009, 18:48   #14  |  Link
thewebchat
Advanced Blogging
 
Join Date: May 2009
Posts: 480
You are introducing a chroma shift with this line:

Code:
## MDegrain causes a chroma shift (yes it does, with pel>1 !)  We compensate by shifting chroma towards plain EDI by a small notch 
##   [ 1+(x-1)/(1+(x/5)^4) ] on 128-centered diff-clip (x-y) ==>  [ x y - abs 2 < y x x y - abs 1 - 1 x y - abs 5 / 4 ^ + / 1 + x y - x y - abs 0.0001 + / * + ? ]
stage2 = stage2.mt_lutxy(edi,"x y - abs 2 < y x x y - abs 1 - 1 x y - abs 5 / 4 ^ + / 1 + x y - x y - abs 0.0001 + / * - ?",Y=2,U=3,V=3)
because this has been addressed upstream in v1.10.2.1

Quote:
* 1.10.2.1 (28.08.2008 by Fizick)

* MVDegrain1-3: fixed old rounding error resulted in some chroma tint (some thanks to Didee for bugreport, but it could be more persevering in official MVTools forum thread
thewebchat is offline   Reply With Quote
Old 2nd November 2009, 20:14   #15  |  Link
Terranigma
*Space Reserved*
 
Terranigma's Avatar
 
Join Date: May 2006
Posts: 953
Thanks for pointing that out, but i'm pretty sure that this chroma shift you're referring to is not the issue bizz & buzz was referring to (since i'm pretty sure he was using the latest version of mvtools 1 which would mean the shifting would have occured), and now that I think about, it's not rfilter either. It's this:

Quote:
Originally Posted by "Fizick
2.4.0 beta (09.01.2009 by Fizick)

* MDegrain: Changed degrain weight method to more strong, some internal rewriting.
In any case, i've reuploaded the "tempgaussmcmvt2" script with the added changes.

Zshare
Attached Files
File Type: rar TempGMCMVT2v2.rar (4.7 KB, 807 views)
Terranigma is offline   Reply With Quote
Old 2nd November 2009, 21:37   #16  |  Link
thewebchat
Advanced Blogging
 
Join Date: May 2009
Posts: 480
It is unrelated to whatever issue bizz&buzz was having, but I still wanted to point that out.
thewebchat is offline   Reply With Quote
Old 2nd November 2009, 23:56   #17  |  Link
Didée
Registered User
 
Join Date: Apr 2002
Location: Germany
Posts: 5,389
Thanks for the pointer, though it's only a reminder. This point already was mentioned earlier.

"Chroma shift" is a slippery formulation here. (Congrat's, if intended.) It's a valid denomination for what is happening, but predominantly it suggests sort of an "incorrectness", like e.g. wrong spatial position of chroma samples, or possibly a shift of hue. Which it is not.

It is not incorrect in any way. It is a bit suboptimal, yes. But it is still correct.
All it does is to take the end result, and adjust each chroma pixel a notch towards the chroma pixel of the initial EDI interpolation. Since the initial EDI interpolation can be assumed to be technically correct, so is the result of this operation. The only loss is a tiny bit of temporal stability, at worst the result will have some minor additional bob shimmer in the chroma channels. So little, and and in such places, that nobody ever would see a difference. Set up a blind test, it'll come in close to 50%.

With the affected older MVTools versions, it was a good thing. And after MVTools's correction, it doesn't matter. TGMC is not pixel-exact / value-exact anyway, and ... wait ... Ooohh, waitwaitwait! Make a PSNR comparison! Didn't try, but I bet that this shift operation increases the PSNR score of the result! Strike!

Leave it in, kick it out, whatever. It doesn't matter.
__________________
- We´re at the beginning of the end of mankind´s childhood -

My little flickr gallery. (Yes indeed, I do have hobbies other than digital video!)
Didée is offline   Reply With Quote
Old 16th February 2010, 04:58   #18  |  Link
Dogway
Registered User
 
Join Date: Nov 2009
Posts: 2,352
Quote:
Originally Posted by Terranigma View Post
His is the MT version. Update the non-MT version i've posted as well. :P
That stands for MaskTools or MultiThreading? That this by now support Multithreading? EDIT: Seems that yes

Also whats the difference between your last version, and the EEDI3 one, besides EEDI3? which one should I use (in quality terms)?

Does TempGauss work for Hybrid should I say(?) like some 3-progressive+2interlaced frames material? or just plain bobbing?

MT=true is outdated? it asks me for MvAnalyseMulti which I think is from an older MVtools?

Sorry for the questions, Im new to this filter...

Last edited by Dogway; 17th February 2010 at 13:53.
Dogway is offline   Reply With Quote
Old 17th February 2010, 14:12   #19  |  Link
hydra3333
Registered User
 
Join Date: Oct 2009
Location: crow-land
Posts: 540
Well, just above the recent post at this link http://forum.doom9.org/showthread.ph...74#post1374374 Didee seems to use tempgaussmc_beta1u which could possibly be the version from 3Nov just a few posts above here (post #15).
hydra3333 is offline   Reply With Quote
Reply

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 21:28.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.