YATTA Manual, A practical approach (Yet Another Telecide Tool for Anime)

[TheRyuu]

The goal was to produce something a little more practical then what exists out there. Eventually I would like to make it html with pictures.

*DISCLAIMER*
If you have no idea about IVTC patterns and what not please don't use yatta. Every time you do a kitten dies. Please stick to tfm().tdecimate() and company if you don't know jack shit about patterns.
*END DISCLAIMER*

Due to character limits:
http://warpsharp.info/yatta.txt

---------
CHANGELOG
---------

2011-06-09: v0.41 Omit junk from prologue I didn't like
2011-06-07: v0.4  Added vfr section
2011-06-07: v0.3  Nice transport stream (constant pattern between commercials)
2011-06-07: v0.2  Mentar's shitty transport stream pattern guidance
2011-06-07: v0.12 Double for real this time, added stuff after PG
2011-06-07: v0.11 For real this time, added dumb shit I forgot
2011-06-06: v0.1  Initial creation of substandard document.

[mastrboy]

Now this is something i have missed, that old yatta.chm help file is not very helpful.

Thank you

[Dogway]

lol it does sound cool eh, I also used it to title an artwork of my own : P
I know nothing about yatta, I gave it a try once but, I might look into it again. Just one question, is it always (if done properly) more accurate by using yatta than tfm+tdecimate or does it depend on source?

[mandarinka]

Well, tivtc by itself won't for example respect pattern reliably. For example with anime, where there are often duplicated frames (but with analogue masters they aren't identical) tivtc will decide on matching (and decimating) just by judging certain metrics. If you use yatta's pattern guidance, you'll have it matched and decimated precisely according to the pulldown pattern. Normal tivtc operation is somewhat prone to mismatching (for example, when there is interalcedly looking stuff onscreen - happens often in anime) and postprocessing false-positives (I encode anime, so the experience comes from that). By enforcing the pattern, you can sometimes prevent misdecimation too...

Aside from that, yatta is a very handy tool for overriding ivtc decisions even if you don't use pattern guidance and just take the decisions collected during the YMC (yatta metric collector) run. It is great for freezeframing (using mouse or hotkeys), variable framerate (24fps / 30fps), and sectioned filtering. It makes those actions much easier/less tiresome to perform.

Last but not least, yatta serves you as a handy tool to look up possible mismatches, interlacing, chroma interlacing and other nasty interlacisms in your source by allowing you to look up and browse frames that reach specified metric threshold, in the gui.

[Dogway]

So by default anime is not as easy to decimate to just selectevery() because pattern changes constantly? or is it the exception?

[mandarinka]

That's not what I meant. Whether the pattern changes or not, anime usually has duplicates (but also close-but-not-really-duplicates). Tivtc does good job (except those errors I mentioned, sometimes, but no computer program is [s]perfect[/s] sentient), but it just doesn't necessarily respect the pulldown pattern. If it can satisfy the metrics checks it does by using field n+2 instead of field n (that should get used according to pattern), it might do exactly that. Or if its decimation criteria are met by frame p whereas according to pattern a same but not identical (noise, rainbows, mpeg2 artifacts) frame p-1 is suppossed to be drop, again it might choose the non-kosher (but usable) choice.

Usually the difference isn't important (unless you are neurotic or really perfectionistic), but sometimes it might. Generaly speaking, as long as tivtc doesn't visibly fail on a frame (completely wrong match, healthy frame garbled by postprocessing), its results will be on the same level as from manual yatta ivtc, I would say.

P.S. Some long-time hardcore user should be able to explain yatta's strengths better...

[TheRyuu]

Originally Posted by Dogway

lol it does sound cool eh, I also used it to title an artwork of my own : P
I know nothing about yatta, I gave it a try once but, I might look into it again. Just one question, is it always (if done properly) more accurate by using yatta than tfm+tdecimate or does it depend on source?

Yatta just provides a means to adjust these matches, so it's really only what you do with it. So if you botch something up in yatta it won't be as good as just tfm().tdecimate() (thus the disclaimer at the top of this thread).

Originally Posted by Dogway

So by default anime is not as easy to decimate to just selectevery() because pattern changes constantly? or is it the exception?

Depends whether it's hard or soft telecine.

[OvejaNegra]

ok, i have a long time fighting with tivtc <> Yatta and i just cant find the usefullness Yatta:
1- What is a "pattern"exactly?
2- "tivtc will decide on matching (and decimating) just by judging certain metrics" <- Whats the problem with that? If you use yatta's pattern guidance, you'll have it matched and decimated precisely according to the pulldown pattern <- And whats the good use of that?
3- "completely wrong match" Not in my experience ¿what settings do you use for matching?
4- "for example, when there is interalcedly looking stuff onscreen" what are you referring exactly?
5- "postprocessing false-positives" mmm, well, not really What settings do you use for post processing? I can give you a formula that works very good with the false positives.


Yes, of course a hand made job is better that a computer decision, but i cant see the porpuse of the time inversion with YATTA vs tivtc with good settings.



Please i'm not attemting to start a flamewar here but simply i dont see WHY so much work with YATTA.
Describe the problems you have with TIVTC and with what anime

(The thing is that i wrote a small tool for TIVTC overriding and i''m just happy with it, but if you can show me the power of YATTA then i will write my Own YATTA

YES, I'm a total perfectionist.
Problems that i hate the most:
Undetected interlaced frame (i have good settings for that)
Jumpy pans because bad duplicate detection (i just tweak the settings of video/dup detection according to the source and apply a denoise filter prior to detection if necesary)

And thanks for this, the Yatta documentation is a little.............bad.

Sorry for my bad english if i say something wierd.
Thanks

[mandarinka]

As for images that I have seen ruined by tfm's matching or PP, see these two examples: #1 #2
I think that gives an idea...

Sadly I couldn't get an image presenting another problem type of frame (sorry, it would take a lot of time to remember & locate some place where it happened to me): mouth-only movements in anime when the mouth is very small. The difference often can't be judged by metric, both good and wrong matches can have the same metric, or the good one can have worse metric for whatever reason. However if the scene is matched according to the pattern, such wrong decissions won't happen (unless the dvd actually violateds the pulldown pattern, let's hope that doesn't happen that often, heh).

Oh, and what is 'pattern'...
Let's take the example from anime again. The animation is generaly created at 24 /23.976/ fps. Howeer they mostly get distributed in ntsc format, so they telecine it (here is where we come to the problem). To get the best result, you generaly want to undo that process as well as you can. So the 'pattern' would be such configuration of filedmatching and frame-decimating (taking in 10 fields and spitting out 4 frames) that would match the pulldown pattern used before (would be the inverse of what the original telecining did).

[OvejaNegra]

Sorry i cant open the pictures righ now, probably the hoster does not loves me....

But i can imagine because i spend a lot of time trying to find the best deinterlacer, please try this

interp = last.nnedi(field=-2)
me=last.tdeint(type=2,edeint=interp,mode=2,full=true)

and set clip2=me of course, use cthresh=0 to test it, you'll see that the frames look 99% like the original, give it a try and give me your opinion.
And use a low cthresh value, like 1 (dont be afraid) or 2, it catches 99.99% of the combing, and in case of false positive, if you use that deinterlacer, you wont notice the difference.

i know the problem of the small mouths you talk about, but:
i tested all the matching modes and the micmatching setting (i'm talking about whole chapters tested allmost frame by frame) and mode =5 +micmatching=1 always find the best match (unless there is a bad edition or something like that) and if the frame cannot be matched the deinterlacer i give you does a good work. Usually those imposible frames are most of the time 5 to 10, no more (in my experience) and the result is perfect (for my eyes) yes, maybe YATTA can produce 100% accurate result (believe me i tested all the possible matches) but i'm talking about 10 frames, not 1000.

Of course my experience is only with 1997after anime, maybe stuff from before is a lot worse.

"unless the dvd actually violateds the pulldown pattern" IT happens, that's why tivtc try to find the best match when possible.

"Oh, and what is 'pattern'" You are telling me that you can use YATTA to declare the matching order and the duplicates, frame by frame? Sorry again, but why?

Find a piece of a vob and proces it with YATTA, and send it to me, ill do the same with Tivtc and will compare screenshots+avs scripts + fixed frames (if any). I have a very bad fragment for you here (it's hentai, not to publish screenshots here).

I'm not trying to questionate your skills with yatta, but simply i still dont see why so much work? Spending a little time tweaking your scrips is a less time inversion and the result is excellent (of course, YATTA can be usefull with REALLY bad sources, but i havent seen one yet)

Thanks,

PD: The Yatta docs sucks!!

[mandarinka]

Heh, maybe you just aren't total perfectionist (since you used the word...)?

Originally Posted by OvejaNegra

"Oh, and what is 'pattern'" You are telling me that you can use YATTA to declare the matching order and the duplicates, frame by frame? Sorry again, but why?

Not frame by frame, but for the whole video, or a defined section of it. Actually, it can be a very fast way to perfectly IVTC a whole video, if the pattern is constant. Often (with hard-telecined footage that hasn't been digitaly edited after pulldown) there are just few pattern switches/shifts throughout.

Originally Posted by OvejaNegra

interp = last.nnedi(field=-2)
me=last.tdeint(type=2,edeint=interp,mode=2,full=true)

Why would I do that when the video is IVTCable? Anime shouldn't be treated with deinterlacers (the only 60i stuff in it are credits, if made with computers). If nothing else, why don't you use nnedi3, btw.

Originally Posted by OvejaNegra

"unless the dvd actually violateds the pulldown pattern" IT happens, that's why tivtc try to find the best match when possible.

Point is that it will hardly happen for a single frame - it will happen for a scene for example, and yatta is more than prepared for that.


P.S.
I'm not suggesting for you to go learn & employ yatta, I was telling you why you or anyone might think about it.

[TheFluff]

If your IVTC filter gives you results that are satisfying to you without manual intervention, then YATTA is not for you. If it does not, then YATTA might be one way to solve the problem.

[OvejaNegra]

Well, actually yes, im very perfectionist, but with a reasonable dosis
"Not frame by frame, but for the whole video, or a defined section of it. Actually, it can be a very fast way to perfectly IVTC a whole video, if the pattern is constant. " I'm trying to find a good tutorial for that, anyone knows?
"there are just few pattern switches/shifts throughout." tivtc cant handle that?
"Why would I do that when the video is IVTCable? Anime shouldn't be treated with deinterlacers (the only 60i stuff in it are credits, if made with computers). If nothing else, why don't you use nnedi3, btw." i use that after every combination of matching can't produce a good result, i wont call it a deinterlacer exactly, but it gives the best results, even a lot better than nnedi3

"Point is that it will hardly happen for a single frame - it will happen for a scene for example, and yatta is more than prepared for that." Why? tivtc always find the best posible match, and not for a single frame of course.


"P.S.
I'm not suggesting for you to go learn & employ yatta, I was telling you why you or anyone might think about it. " Of course my friend, but if you could cut a small piece of a problematic vob and put a small info here of how process it with YATTA, so i could see the magic,
please


"If your IVTC filter gives you results that are satisfying to you without manual intervention, then YATTA is not for you. If it does not, then YATTA might be one way to solve the problem. " YES, i totally agree with you, but i need a quick 1 2 3 course with YATTA to make a test and see for myself. Everybody says that YATTA is nirvana and actually nobody tells me: Here, take this piece of vob and do this with tivtc.... see? it'cant be fixed, now do this and this with YATTA... chachan!! perfect! <- that's what im looking for.

Maybe i just had luck with my sources and everything was ok with tivtc and i don't have a really bad DVD.

[mandarinka]

Get the tutorial OP offers in this thread?

Originally Posted by OvejaNegra

tivtc always find the best posible match, and not for a single frame of course.

Except it doesn't *always*. See my mouthmoves complain earlier. Tivtc will match fields based on metric criteria - whatever suits best is used, and that's why it can ocassionaly fail, when the best metric isn't actually the good choice. If you just followed the pattern (yatta~), the proper decission would be taken, though. /And no, micm doesn't help here./

[TheFluff]

Originally Posted by OvejaNegra

"If your IVTC filter gives you results that are satisfying to you without manual intervention, then YATTA is not for you. If it does not, then YATTA might be one way to solve the problem. " YES, i totally agree with you, but i need a quick 1 2 3 course with YATTA to make a test and see for myself. Everybody says that YATTA is nirvana and actually nobody tells me: Here, take this piece of vob and do this with tivtc.... see? it'cant be fixed, now do this and this with YATTA... chachan!! perfect! <- that's what im looking for.

Maybe i just had luck with my sources and everything was ok with tivtc and i don't have a really bad DVD.

The forum has a quote button, please learn to use it. Your posts are really hard to read.

YATTA is definitely not "nirvana" or a magic solution to anything. It's a tool that was created because some people wanted a less suicide-inducing way to hand-edit Decomb override files. It's still like that; if you really are masochistic enough to want to mess with your IVTC matches manually, YATTA makes it slightly less horrible to do so than if you were using a plain text editor. If you have not yet encountered a DVD or TS that is bad enough for you to want to do this, you should probably not be trying YATTA out.

Pattern guidance is highly overrated, in my opinion, but some people swear by it because it lets them "know" that the entire video is perfectly IVTC'd down to the last frame. Of course, in 90% of the cases TIVTC will do effectively the same thing, but they like their placebo.

YATTA also doubles as an Avisynth script generator/editor with a few convenience functions AvsP lacks, but that's really not the point of it.

[OvejaNegra]

Get the tutorial OP offers in this thread?

sorry, yes i just forgot it!! i will read it.

The forum has a quote button, please learn to use it. Your posts are really hard to read.

OK

[kedautinh12]

Originally Posted by TheRyuu

The goal was to produce something a little more practical then what exists out there. Eventually I would like to make it html with pictures.

*DISCLAIMER*
If you have no idea about IVTC patterns and what not please don't use yatta. Every time you do a kitten dies. Please stick to tfm().tdecimate() and company if you don't know jack shit about patterns.
*END DISCLAIMER*

Due to character limits:
http://warpsharp.info/yatta.txt

---------
CHANGELOG
---------

2011-06-09: v0.41 Omit junk from prologue I didn't like
2011-06-07: v0.4  Added vfr section
2011-06-07: v0.3  Nice transport stream (constant pattern between commercials)
2011-06-07: v0.2  Mentar's shitty transport stream pattern guidance
2011-06-07: v0.12 Double for real this time, added stuff after PG
2011-06-07: v0.11 For real this time, added dumb shit I forgot
2011-06-06: v0.1  Initial creation of substandard document.

Link died, can anyone got yatta?? Please share me. Thanks

[Reel.Deel]

Originally Posted by kedautinh12

Link died, can anyone got yatta?? Please share me. Thanks

archive.org is your friend: https://web.archive.org/web/20160610...info/yatta.txt

[Emulgator]

http://ivtc.org/