Ultimate IVTC settings?

[DJ Bobo]

Hi everybody!
I've been transcoding Anime for a long while now. I tested GreedyHMA, Decomb & IVTC 2.2

I've found that IVTC 2.2 delivers the best results with the following settings:
ivtc(32,11,110)

That gives so far smooth and clean output of every difficult scene I've seen til now.

I want to ask you to test these settings and see if they work universally on any hard scene you know. If it doesn't work properly, please try to find better settings, or give me an alternative setting in another ivtc plugin.

thank you!

[Guest]

IVTC22 fails on (for example) "Tenchi The Movie 2". It also fails on much of the Cowboy Bebop material.

Decomb is also having trouble with Tenchi (as are all other IVTC programs). I hope to improve its behavior in the next release.

[DJ Bobo]

Havn't you find good settings for those two?

I mean, tweaking the different settings helps much, my goal is to find a setting that works perfectly with any source.

[Guest]

>my goal is to find a setting that works perfectly with any source

If there were such settings, don't you think the program author would have made them the immutable defaults? :)

[DJ Bobo]

That's excatly the point: with the recommended settings in the help file, I get almost in every Anime some jerky output. And I kept optimizing the parameters and I found til yet, that (32,11,110) work very good (a lot better than the recommended settings, "perfect" may be!), that's why I was asking if there are scenes where IVTC fails with this parameters, in hope to optimize further if it's possible or if it is needed ;)

[Milkman Dan]

Originally posted by neuron2
IVTC22 fails on (for example) "Tenchi The Movie 2". It also fails on much of the Cowboy Bebop material.

Decomb is also having trouble with Tenchi (as are all other IVTC programs). I hope to improve its behavior in the next release.

Decomb did a pretty good job on TMiL2, as I just ripped it the other day. Granted, there is still some artifacting around mouths and such at times, but considering the 'telecining' is awful to begin with, I think it did rather admirably.

For the Tenchi OVA's, eps' 1-6, Decomb does fine with the default settings, as does IVTC 2.2. Episodes 6 and 7 have a screwed field order, which Decomb handles by specifiying reversed fields. IVTC2.2 handles it just fine with the default settings. 8-13 are fine so far as I've done them.


But hey, if you're lookin' to improve the output, I'm certainly not going to complain. ; )
I own TMiL 1 and 2, and the OVA's on DVD, so if you need me to run tests on unreleased builds, feel free to drop me an email. Should be in my profile.

EDIT: Nevermind. This isn't one of those forums. email is jhickman at parallaxmedia dot net

[DJ Bobo]

@ Milkman Dan
Perhaps you misunderstood me, I just want to find the right parameters, since I can't change the code (I'm no programmer).
With the default settings of IVTC, I got almost in every anime some bad output. So I adapted the parameters every time, and I've come til now to (32,11,110) that gives me very nice output with the difficult scenes I've seen til now (I'm collecting them ;))

You may test those parameters with difficult scenes in Tenchi Movie 2 und reply here if you want. You may try to find better parameters if it doesn't work properly.

[Milkman Dan]

Erm...I wasn't replying to you, I was replying to neuron2, and pretty much exclusively to what he said...

[OUTPinged_]

@bobotns:

what was ivtc2.2 again? wasnt that old Wiz's IVTC routine used as default in gknot?

can you list here some of the anime sources you think that setting will work best for (dvd region included)?

(mmm, i am sceptical about that post, i have several decomb settings for several sources i have (which i believe work best). i have some source where you gotta make a field drop to make things work smooth(no avisynth dlls, tmpeg only and tons of tweaking). and here comes that guy and claims he found an ultimate setting for each anime out there using old tool.)

[manono]

Hi OUTPinged_

Where you been hiding?

Yeah-IVTC 2.2 is from The Wizard. His version 2.1 is the one included in GKnot.

[DJ Bobo]

It's not the same routine as in GKnot. GKnot uses version 1.x or something, which is slow and brings lot of shit.

I tested IVTC 2.2 with parameters (32,11,110) on:
1) Some of the Dragon Ball Z US-DVDs
2) the US-DVDs of "Arc The Lad"

I have a lot of other Anime but they were made with worse settings or with other ivtc plugins like GreedyHMA and Decomb.

I'm going to test the newest parameters on the HK-DVDs of Saint Seiya in the next days.

[manono]

Hi-

According to the Inverse Telecine Readme found in the Docs folder with GKnot, it's version 2.1. Maybe 2.2 is a big step forward.

[OUTPinged_]

bah.

at least the one in gknot didnt feature smart deinterlace of frames detected as interlaced.

i have tested 2.1(or was it 2.0) myself and it left me a pretty bad memory. no deinterlacing, no guidance, dumb field selection algo.

cmon, it didnt ever detect film content properly on scenechanges for half of anime dvds i have got. i aint talking about pure 30fps sequences at all.

how old is that 2.2 version anyway?

PS> it doesnt even begin with "de", end with "mb" and have "co" in the middle. now will you believe me ? =)

\o/

mamono, i may be wrong for that guidance thresh after all. with clear source it seems you can go up to 90 and be pretty safe, well maybe a couple of interlaces at gay scenechanges will slip through.

[DJ Bobo]

@ Outpinged
Changing the parameters helps a lot, it blows decomb away! with the settings (32,11,110) I get fantastic results and it's faster than Decomb!
I tested also the new parameters with the first episode of Saint Seiya, absolutely fantastic results!

If I find a scene, where those parameters don't catch, I'll report it here.
I'm collecting the difficult scenes so that I can find a universal setting that works with all of them!

[Milkman Dan]

In my experience, Decomb was a great deal faster than IVTC 2.2. About 50%, in fact.

That was on my K6-3 though. Perhaps it's an optimization thing.

All things equal though, I see Decomb as a more versatile solution. And it's actively developed. Wizard is poking around again though, and is talking with neuron2, so perhaps we'll see some code integration. Until then though, I'm sticking with Decomb.

[DJ Bobo]

Almost 2 months now and it comes now to an end: my settings (32,11,110) aren't universal anymore :D
I found an Anime that doesn't ivtc well with my old settings, because it is really bad telecined (captured from LaserDisc or something).

BUT!
Here are the new settings that I found and that work well with that problematic anime and with my collected "difficult to IVTC" videos:

ivtc(60,11,110)

As always using the old good IVTC 2.2 (you find this on the Doom9 download page)

Please try this brand new setting and tell me if it works well with your favorite problematic anime.

[Kyo]

I try to encode the FLCL DVD(anime) 01 and:

1) IVCT22 nop
2) GreedyHMA not full tested but nop
3) with decomb I use:

Telecide(Post=true,Blend=False,Guide=1,Gthresh=44,Threshold=20,Chroma=True,dthreshold=12)

When use the default "Telecide()" decomb deinterlace some good frames and this is very noticeable, also the motion is not so smooth, when I use guide=1 the smooth is much better(perfect) but in some escenes the guidance is not accurate, the why for Gthresh=44, perhaps when I raise Gthresh escenes with small interlace (some mouths) pass the filter also if I use Post=true.

The question is: anyone have any sugerence for this dvd or other ivtc method (I really don`t want do it with Tmpeg manually :o )

and excuse the bad english/grammar/errors ;)

Thanks

[OUTPinged_]

Almost 2 months now and it comes now to an end: my settings (32,11,110) aren't universal anymore

*sigh*

they were not from the very beginning.

there are no "universal" settings. how many dvds have you tested that setting with? 3-5? test it with 100 anime dvds and only then claim it "universal".

[DJ Bobo]

@ outpinged
Let me count for you: about 70 DVDs! the last one didn't work with (32,11,110), so I tweaked it again to (60,11,100).
Please don't bother me anymore, you have always something to say when I post my settings.
If you wanna help to find better settings than mine, you're welcome. If not, please don't look into this thread.

[RadicalEd]

uum.. bobotns, I tried these settings for the first time with a DVD that was giving me a lot of interlaced mouths no matter what settings I used in decomb. Anyway Vampire Princess Miyu R1 is no good with these settings. I'd find better ones.. but I think I'll just stick with decomb, ya know?

[DJ Bobo]

@ RadicalEd
I'm concentrated on IVTC 2.2, but if you find settings in Decomb that work better than ivtc(60,11,100), just post them, if they work well, I'll send IVTC 2.2 to hell!

[OUTPinged_]

Kyo:

what if you will try

Telecide(Post=true,Blend=true,Guide=1,Gthresh=25,Threshold=25,Chroma=True,dthreshold=10)

check it and report here, is it jerky/leaves interlaces/etc.

[Kyo]

Thanks OUTPinged_ for the reply i'm going to try NOW :D !
and post late :p

at the moment I'm testing to do a segmented-ivtc and the script look
like:

dvd=mpeg2source("K:\flcl\Flcl01.d2v")
part1=Trim(dvd,0,2941).Telecide(Reverse=True,Post=True,Blend=False,Guide=1,Gthresh=5,Threshold=60,Chroma=True,dthreshold=9).Decimate(cycle=5).BicubicResize(592,448,0,0.65,4,0,712,480)
part2=Trim(dvd,2942,3840).Telecide(Reverse=True,Post=False,Blend=False,Guide=1,Gthresh=46,Threshold=60,Chroma=True,dthreshold=9).Decimate(cycle=5).BicubicResize(592,448,0,0.65,4,0,712,480)
part3=Trim(dvd,3841,39277).Telecide(Reverse=True,Post=True,Blend=False,Guide=1,Gthresh=5,Threshold=60,Chroma=True,dthreshold=6).Decimate(cycle=5).BicubicResize(592,448,0,0.65,4,0,712,480)
ending=Trim(dvd,39278,0).Telecide(Reverse=True,Post=True,Blend=False,Guide=1,Gthresh=5,Threshold=60,Chroma=True,dthreshold=6).Decimate(cycle=5).BicubicResize(592,448,0,0.45,4,0,712,480)
Return part1+part2+part3+ending

But in some escenes the interlacing on mouths (especially on diagonal-centered mouths) is not gone :(

this is the first time that I evaluate the power of Avisynth :D
and reduce the sharp on ending (0.45) seem a god idea to save some bitrate. well later post a conclusion!

/excuse the English or could be the grammar or ...

[OUTPinged_]

Try to look if mouth problem appears for you with 100 gthresh or 0.

and a hint: it is a bad idea to save bitrate on resizing. FLCL does compress very well anyway.

[Awatef]

So people, that's it! I found the ultimate IVTC settings using IVTC2.2:

ivtc(44,11,95)

I think, it can't go any better than that, using this filter.
It even works with sources like the Evangelion US DVDs with over 80% success rate.
For such DVDs, you *may* use FieldDeinterlace() after ivtc to get rid of the remaining interlacing artifacts. But keep in mind that this might make motion a little jerky, since in DVDs like those of Evangelion, even the frames supposed to be normal have interlacing artifacts (because of DVD authoring error), that FieldDeinterlace() will also blend -> you get ghosting instead of horizontal lines in normal frames -> it's like driving a car with handbrake on. So it's your choice:
[80% clean + 20% interlaced] and 100% smooth motion
[100% clean] but only 80% smooth motion

For other normal DVDs, that are properly telecined, you don't have to use FieldDeinterlace, ivtc(44,11,95) alone works like a charm (99% success rate)

And just to get OUTPinged_ out of my way: yes, I tested this setting extensively on many various Anime DVDs.

[FiW]

hmm.. i'am use ONLY doubleweave() & pulldown().

example: 'pearl harbor'

loadplugin("c:\progra~1\gordia~1\mpeg2dec.dll")
mpeg2source("g:\.dvdivx\.pearl.harbor\pearl.harbor.d2v")
p1 = trim(0,102601).doubleweave().pulldown(0,2)
p2 = trim(102602,252592).doubleweave().pulldown(0,2)
p3 = trim(252593,0).doubleweave().pulldown(0,2)
p1+p2+p3
crop(7,55,703,365)
bilinearresize(672,288)

example: 'requiem for a dream'

loadplugin("c:\progra~1\gordia~1\mpeg2dec.dll")
mpeg2source("g:\.dvdivx\.requiem.for.a.dream\requiem.for.a.dream.d2v")
p1 = trim(0,522).doubleweave().pulldown(1,3)
p2 = trim(523,5007).doubleweave().pulldown(1,3)
p3 = trim(5008,88179).doubleweave().pulldown(0,2)
p4 = trim(88180,0).doubleweave().pulldown(0,2)
p1+p2+p3+p4
crop(5,15,713,447)
bilinearresize(464,240)

IMHO, it's best method...

[Awatef]

@ FiW
Well, I'm impressed! I tested doubleweave().pulldown(0,2) on some short 2~3 minute anime clips and your setting is working like a charm.
But I'm not so sure, that it will work like a charm for a whole anime episode or movie.
If I'm not wrong, you're working like the manual method of TMPG, changing the pulldown setting, when the pattern changes, am I right?
But knowing that Anime changes pattern much much more often, your method will be like hell (are you sure your movies weren't film content anyway?!)
I'm not into encoding a whole anime disc right now, may be I'll have time next week and than I'll see how far your method will go! (probably not so far ;))

[FiW]

@Awatef

yes, this method similars manual ivtc in tmpgenc. use this method in avisynth better usability & more fastest (for create).

[Awatef]

Well, if it so, no need to test your method extensively, I already have a conclusion:
1) your method is perfect for people having time & nerves, to change pattern every few minutes, if not seconds! (talking about anime)
=> Quality: A+ / Speed: D- before encoding & A+ while encoding
2) my method is more suitable for people with less time/nerves, and the result will be very near to what one gets using your method.
=> Quality: A or B with supposed "impossible to IVTC" video / Speed: B or C if FieldDeinterlace() is used

[The Link]

@ Awatef
... that FieldDeinterlace() will also blend -> you get ghosting instead of horizontal lines in normal frames I´m not an expert, but I think you can use "FieldDeinterlace(blend=false)" so there´s no blending any more. Correct me if I´m wrong!

Regards,

The Link

[Awatef]

@ The Link
Oh, of course! you're right! I completely forgot about it! gomé! (=sorry! :D)

[Billy|B-A]

A few quick questions.

What do the values of each of the ivtc settings mean? (x,xx,x)
What are their respective greatest/least values?
What are the rythem (cant think of a better word) of the values? i.e odd #'s only, multiplied by 8, or even #'s only.

I looked into the readme of IVTC 2.1 and didnt c a answer.

Thanks in advance

[DJ Bobo]

Download IVTC 2.2, in its help file you'll find what you're looking for.

[DJ Bobo]

Well, it's time to post the new settings I guess :D

ivtc(70,10,25)

are the new settings. They are quite untested (tested on only 7 DVDs this far) but they work definitely better: the pattern detection is better, that means the filter adapts immediately to pattern changes. Until now, I havn't encountered any scenes where it "hesitates" finding the right pattern.

As for problematic sources like evangelion, no need for decomb anymore, as it tends to "forget" some interlaced frames, so this is the new command:

ivtc(70,10,25)
BlendFields()

Note that those lines must come directly after the mpeg2source line.
Keep in mind that you won't ever get perfect results with such problematic sources. You've got to live with what you get ;)

As Deinterlacing (here using blendfields) blurs the image a bit, you may wanna follow your script with a sharpening line. I'll recommend UNFILTER from trbarry.

NB: I'm not so sure about what I'm going to say, but I have a feeling that IVTC doesn't work well with newer avisynth versions. Go back to older avisynth versions (1.05 for example) if you experience bad divx5 encoding results (notably the keyframe delay)

[RadicalEd]

awesome, better ways to ivtc stuff like Evangelion is always good.
Cheers for the discovery :D lets hope the avisynth thing isnt too much of a problem


whoa, wait a sec, I just re-read the thread and dj bobo = bobotns :eek: :confused: howd you get your name changed like that?
I never knew :\

[ArdenDag]

@DJ Bobo

Have you tried decomb's updated 'agressive field matching' that Don re-enabled in his latest release? (4.02, I think, not available on his website yet)

Telecide(agg=true)

Also, the noise threshold, before probably an unknown or unchangable variable is also now 'changable', nt is the variable, 100 is default.

I've noticed aggresive field matching works better for bad telecined material (Hong Kong bootlegs), you might want to check his posts and see if that helps :)

Always like seeing things like this come up :)

EDIT: for stupid rephrasing :p

[OUTPinged_]

> Ultimate IVTC settings?
I've found that IVTC 2.2 delivers the best results with the following settings:
ivtc(32,11,110)

BUT!
Here are the new settings that I found and that work well with that problematic anime and with my collected "difficult to IVTC" videos:
ivtc(60,11,110)

So people, that's it! I found the ultimate IVTC settings using IVTC2.2:
ivtc(44,11,95)

Well, it's time to post the new settings I guess
ivtc(70,10,25)

Posting new "ultimate and absolute best settings" every couple of months, eh? ;-) What's up with these people anyway?

ivtc(70,10,25)
BlendFields()

Well i could tolerate your "ultimate settings" for that old thing (it's your choice if you want to have interlacing on scenechanges instead of blended frames), but that does it.

The guy that chooses BlendFields over FieldDeinterlace can't be trusted. Either he can't see the difference, or he dislikes DG's tools, or he never bothered to actually check it.

@bobotns: Yea, i know you asked me to not reply, but you guys are funny :-)

[DJ Bobo]

@ ArdenDag
I'll test it next time I get bad telecined things :D

@ _outpinged
There is no more interlacing on scenes changes thanks to the new settings. I was like 12 hours testing until I found the way to get perfect scene changes (you can't imagine how happy I was to finally find those settings! :))
It is better to use blendfields instead of fielddeinterlace, because blendfields deinterlaces all frames no matter what. But fielddeinterlace switches all the time between "skip" and "deinterlace", so you'll see a kind of pumping in the picture, it keeps switching between "sharp" and "less sharp" and that's indeed quite annoying. So it is better to deinterlace all with blendfields, and add unfilter if you feel like sharpening a bit.

NB: this is not a re-invitation to this thread. Please just keep watching and don't reply! everytime you reply I feel a rage going through my body! :o

BTW, perfection doesn't exist in this world. It's all a matter of "best possible", so there is always a possibility to improve ;)

[ArdenDag]

@DJ bobo

FieldDeinterlace(full=true,dthreshold=0) (or 255, I forget) should deinterlace all frames, whether they're combed or not.

[Guest]

Let's keep things impersonal, please, gentlepeople. Thank you! Users should not try to impose personal protections on threads.

Where do I get the Avisynth 2.5 version of IVTC22? I assume it supports YV12.

'Ultimate' seems a bit of a stretch. Like any other IVTC software, IVTC22 succeeds brilliantly in many cases while failing others. My testing showed it to fail on my 3rd Decomb torture file.

[RadicalEd]

I'm still wondering how bobotns got his name changed to dj bobo like that :\

[Guest]

Originally posted by RadicalEd
whoa, wait a sec, I just re-read the thread and dj bobo = bobotns :eek: :confused: howd you get your name changed like that?
Maybe he has 2 logins.

[DJ Bobo]

@ neuron2
It's obvious that there still some "torture" scenes as you call them where the filter will fail.
For example, during my testing, I got some really weird video part, with bad telecining. I can change the settings to whatever I want, I won't ever get it clean. With some settings, the first scene is ok, the second is bad. With some other settings, the first scene is bad, the second is ok. I also noticed that the first frame the clip starts with has a big influence on how the IVTC filter will work later! if I start one frame later or earlier, I got often different results! (not necessarely of course)

But on normally telecined videos, I can swear, despite limited testing, the new settings deliver better output, especially on the old weak point, the scene changes ;)
I'm very proud of those new parameters :D

[Rhaegar Targaryen]

Oh this is just a ridiculous thread. And Outpinged (sp) hit it on the head when he pointed out that the guy who claimed his values were "ultimate" provided no less than 5 different so-called "ultimate" values... each one more "ultimate" than the other (?). And his credibility went down the drain when he started defending the use of BlendFields() over FieldDeinterlace().

Of course there are certain scenes that IVTC 2.2 handled perfectly, where Decomb 4.01 (haven't tried 4.05 yet) failed on. However I think it is the EXCEPTION and not the NORM for IVTC 2.2 to suceed where Decomb 4.01 fails! So please get your toolset up-to-date, and teach yourself all of the various options for Decomb (and try them) before going out in a public forum and blathering nonsense about how you have found 5 different "ultimate" IVTC settings (!) for an IVTC plugin that hasn't been developed in a year or more!

[Rhaegar Targaryen]

Originally posted by neuron2
Where do I get the Avisynth 2.5 version of IVTC22? I assume it supports YV12.

Hey Don. I don't think there is any YV12 (AVS 2.5) version of IVTC22. As I mentioned above, I don't think it's been developed on in at least a year. I don't even think the sourceforge-based Avisynth 2.0x series had begun when IVTC 2.2 was released.

[DJ Bobo]

@ Targayren
You are also not wished on this thread too :p
You wanna be constructive, welcome, if not, don't even take a look.
You are the one acting ridiculous here, and I know that thanks god, there is many people that consider my advices, because I'm a serious poster, and want that people here profit from every discovery I make!

I know what I'm talking about, just to let you know, until today I ripped over 100 telecined DVDs, I'm not making jokes here, I provide solutions that work, based on real long experience, and as said before, there is no perfect solution, there is *ALWAYS* room for improvement! and please consider that between one setting and another, there is not one day or two, but weeks or even months!
And if you havn't known, ultimate was never a synonym to perfect, so what's your problem?!

I can not afford to work with parameters that don't work universally, since I rip so much, I want a parameter, that works with closed eyes! and when I find a scene which requires improvement, I post the improvement and don't keep it between my 4 walls, just because people like you will complain about updating so "often"! (although it's not tha~t often!)

BTW, I switched to IVTC because Decomb failed in too many scenes where IVTC succeeded, not the contrary :p
And I keep trying almost every new version of Decomb, just to find that IVTC is still the best!
And you've got to read my previous posts, to see why I prefer BlendFields() over fielddeinterlace() on bad telecined video (although arden posted a setting that makes exactly the same like blendfields, so it doesn't make much difference then). On normal telecined video, I don't even bother deinterlacing :devil:

[EDIT]
I just looked at the old threads, the (44,11,95) was posted by Awatef the 22nd of june. I posted the new settings (70,10,25) just 4 days ago, these are 6 months!

[Guest]

I asked you guys to stay impersonal and you are not. Strikes will follow any further occurrences.

[ArdenDag]

@DJ bobo

Did you get around to testing the new version of decomb with aggressive field matching, and a configurable noise threshold? I suggest you take a look, try the simple aggressive field matching on some of that bad material you have, see how it does :)

[DJ Bobo]

@ neuron2
Yeah right, now I'll be a victim of a strike just because I defend my position...

I just tried to delete this damn thread, but havn't succeeded for some reason, please delete it, I don't have the nerves to take any more of this and to confront people that don't have any respect for my work.

I'll just keep my discoveries between my 4 walls then, don't need any of this. Want to serve people, and then this, no thanks!

[manono]

Hi DJ Bobo-

I'm inclined to agree with you that if there are any strikes to be handed out that RT should get the first one for his personal attack. However, you did come back a little too strongly (twice) after being politely warned by neuron2.

And I think you're going to have some problems convincing the people on this forum (myself included) that deinterlacing the whole thing with BlendFields is the answer for a few interlaced frames that slip by whichever IVTC you use.

But I always welcome your posts (even though I might not always agree, as you know) because they are backed up by many tests and a lot of knowledge. So don't give up on us yet. Besides, it's Christmas day, and I'm sure a full pardon will be offered for all crimes. :) Merry Christmas DJ Bobo and all others that pass through here.