Problem when trying to IVTC "Robot Chicken S1" DVDs

[Chainmax]

I am trying to make a standalone compatible Xvid rip of this boxset I purchased a few months ago. The problem is that it has so little motion that it's extremely hard to see any combing and determine field order. Also, afte IVTCing, there's still some residual combing and aliasing which is hard to fix without messing up other parts.
Therefore, I turn to you for suggestions. Here are a few samples:


Residual combing + aliasing:
http://www.bestsharing.com/files/gwWEA264769/Clip1.demuxed.m2v.html

One of the very few scenes where combing is noticeable:
http://www.bestsharing.com/files/Fsai2M264783/Clip2.demuxed.m2v.html

A scene that becomes very screwed up if doing antialiasing:
http://www.bestsharing.com/files/t0mqU8d264792/Clip3.demuxed.m2v.html

[Mug Funky]

if it's anything like the other stuff coming from cartoon network, you might be looking at animation in 30p. i've seen a little of the brak show and ATHF, and those are mainly 30p.

so perhaps you would do better to bob+decimate.

[Chainmax]

What about the combing artifacts in clip2 then, how could they be in there?

[ChiDragon]

Step through the second clip with Bob(), there's some weird blending/warping going on. I'll leave it to someone else to suggest ways to undo this as I have no experience with it.

[Chainmax]

Anyone else then?

[Mug Funky]

looks like speed changes to me.

i'd say just do it at 30p and run an unblender on it to get the most out of the 60 fields (to try keep the blends and blurs to a minimum). so maybe try see if restore24 can be used to go from 60p to 30p.

this judgement is based wholly on clip2 though, but the mocomped speed changes are pretty much unmistakable.

[Chainmax]

So, it would be something like:

a2=Yadif(mode=1)
b2=TDeint(mode=1,type=1,chroma=true,tryweave=true)
restore24(a2, b2)
Blend removal

, right? As for blend removing, which one would you recommend me and where can I download it (some time ago, I looked for MRestore and CDeBlend and couldn't find them)?


Also, could you please take a look at clips one and three? The former shows some residual combing and also aliasing, while the latter becomes quite messed up if using antialiasing.

[Mug Funky]

well, i meant trying to see if restore24 could be set with num and den equal to each other [edit] - i just realised how stupid that statement is, so it doesn't change the framerate at all. the unblending would be done within restore24.

but i've never used it that way so i can't be sure if it'd work.

in the meantime, here's a cludge i put together from another script i had lying around. works pretty well for clip2.


function blendkill (clip c)
{

c.yadif(1)
global clippidoo=last
global edge=clippidoo.mt_edge(mode="prewitt",thy1=0,thy2=255,y=3,u=-128,v=-128).levels(0,.25,255,0,255,coring=false)
scriptclip(clippidoo.selectevery(2,0),"""
\ edge.selectevery(2,0).averageluma() < edge.selectevery(2,-1).averageluma() &&
\ edge.selectevery(2,0).averageluma() < edge.selectevery(2,1).averageluma() ?
\ clippidoo.selectevery(2,0).ydifferencefromprevious() > clippidoo.selectevery(2,0).ydifferencetonext() ?
\ clippidoo.selectevery(2,1) : clippidoo.selectevery(2,-1) : clippidoo.selectevery(2,0) """)
}

mpeg2source("Clip2.demuxed.d2v")

assumetff()

blendkill()


it should do pretty much the same as "unblend", but it works off 60p input and outputs 30p (or 50p to 25p, or whatever). you can replace the yadif line with anything else - i just put it there because it's my current favourite, but tdeint should do a better job on account of your mostly progressive source.

[Chainmax]

What kind of leftover artifacts should be expected from using it? Also, would there be ways to improve on it other than using a better deinterlacer?

[setarip_old]

Hi!

As an experiment (as I've suggested inposts in other threads), you might want to try using MPEGMediator to do the same conversion(s), coinciding with your uploaded example(s)...

[Mug Funky]

@ chainmax:

2 types of leftover artefacts - deinterlacer related (ie. shimmering lines), and bad matches (ie. leftover blended frames). none are really tragic, and it should look completely fine when played back.

so obviously, the best ways to improve it would be:

- using a safer deinterlacer, or
- writing better metrics and using better logic.

i've basically ripped the blend judgment logic from didee's original documentation of restore24, and simplified it a lot. if you break it down, all the scriptclip line is doing is choosing to keep either the current frame, or either the previous or next frames, depending on which one appears "less blended". the amount of blending is judged by how bright an edge-masked version of the input is and whether it sits between the brightnesses of the adjacent frames. i've actually posted that same code fragment with modifications several times before, so if it looks familiar that's why :)

so good luck with it.

[Chainmax]

Ok, when the newest version of TDeint with smart thresholding comes out I'll try your method. Thanks for the suggestion :). One thing though, I once tried to just IVTC and the interkaced frames from Clip2 didn't show up. The only reason I didn't encode it then was because of the leftover combing and aliasing, but I didn't know about Vinverse then. In any case, could you tell me for future rips/caps how to distinguish this kind of scenario (30p animation), along with what are these mocomped speed changes you speak about and how to spot those too?


setarip_old: could you point me to a few of those threads so I can read more about your method?


[edit]Mug Funky, you should post that function in the Development subforum so that other developers can take a look at it.

[Mug Funky]

function is too simple for the devs :)

as for identifying this footage, all i can say is step through bobbed frames and look for obvious warping (including blends where an object in the previous frame sort of "shrinks" into the background, like some of the moving arms do in clip2). that stuff's a sure sign of a motion-compensated slowdown/speedup.

however, this could be scene-by-scene - speed changes happen all the time in editing, especially short-form stuff (music videos, ads) or low budget stuff.

[Chainmax]

I see, thanks. As for determining if it's 30p, it's just a matter of seeing if the interlacing only happens sparsely and for very short periods of time, right?

[Chainmax]

Actually, IVTCing gets rid of the interlacing artifacts from Clip2. It only leaves aliasing and some residual combing which is visible (for instance) on Clip1, but maybe Vinverse can get rid of those. Are you sure it's 30p? How can it be distinguished from the more conventional "24p"?

[Chainmax]

Mug Funky?

[Chainmax]

Last bump...

[Guest]

Bumping like that is frowned upon and several mods consider it a rule 11 violation. Please don't do it anymore. Thank you.

[MOmonster]

Sounds like DEFT or something similar. If you still need help, please upload the clips ones more (I can´t download them).

[Chainmax]

Will do ASAP, thanks :).

[Chainmax]

Ok, here are two clips. The first one shows the issues, the second one displays noticeable distortion when using antialiasing:

[links removed]


Once you download them, let me know so I can delete them.

[MOmonster]

I downloaded them, you can remove them, thanks.

It´s really a strange source with a lot of framerates inside:
30fps, 24fps, 15fps, 12fps...
In your first sample the blends are surrounded by clear frames with dups. I didn´t find this pattern:
c b c b c b ...
only this one:
c d d b c d b c d d ...
Tfm can catch this:
deint = tdeint(mi=66,cthresh=8)
tfm(slow=2,mode=3,ubsco=false,cthresh=8,mi=66,clip2=deint)
I´m not sure if all parameters are right. You should tweak them for your own needs (mthresh,cthresh,mi).
But if it comes to this pattern: c b c b ... anywhere in the source I recommed this:
tdeint(mode=1,cthresh=8)
Mrestore(numr=1001,denm=2002)
I don´t see any annoying aliasing in your second sample, so why using an anti aliasing function on it.
At the beginning of the first sample, there is a short sequenz I would use anti aliasing, but not on the whole source.
Also I need some information about the anti aliasing function you use to give some advice.

[Chainmax]

So, do you think I should leave this as 29.97fps then? If so, should your two suggestions be combined like this:

deint = tdeint(mi=66,cthresh=8)
tfm(slow=2,mode=3,ubsco=false,cthresh=8,mi=66,clip2=deint)
Mrestore(numr=1001,denm=2002)

? As for the aliasing, it's somewhat noticeable in the part where Wonder Woman kicks the washing machine's door. What I usually use for antialiasing is two consecutive EEDI2 calls followed by a downsizing to the original size, and if that's not enough I use the AAA() function posted by Soulhunter a long time ago.

[MOmonster]

No, your script makes no sense, because it will output ~15fps.
This:deint = tdeint(mi=66,cthresh=8)
tfm(slow=2,mode=3,ubsco=false,cthresh=8,mi=66,clip2=deint)
worked fine with your two samples, but if there is a scene in your source this script still outputes annoying blends, and only if this is the case, you should use this script instead:tdeint(mode=1,cthresh=8,tryweave=true)
Mrestore(numr=1001,denm=2002)
Did you allready test the script? Most fields are weaved, so the most frames doesn´t have any noticable aliasing. Also what aliasing produce the artefacts?
I´ll test them soon.

Edit: Yes, because the max real framerate is 29.97fps I recommed to keep this.

Edit2: Ok now I know what is your problem with the anti aliasing. it makes the fine structure just too soft. I´ll have a look on it, but I don´t know if I´ll find a solution.