Deinterlacing problem or Source problem ?

[ncahammer]

I have this source

http://www.speedyshare.com/335905731.html

I use a simple TFM().TDecimate() and I get an excessive trembling, notice the cards shots (sometimes it's hard to say whether is "9" or "8").

am I using the correct IVTC commands ?
can I do anything to stop that trembling ?

TIA

[thetoof]

Maybe that's because you need a bit more than a simple tfm.tdecimate ;)
Try animeivtc(mode=1,aa=2)

[ncahammer]

No, AnimeIVTC actually gives worse results (I did test several modes, none of them produced anything good)
btw, The source is not anime or cartoon, it's a clip from the NTSC DVD release of "Space:Above and Beyond"

[Nightshiver]

And you should heed thetoof's suggestion as he probably knows more about IVTC'ing than you do. animeivtc isn't only for use on anime. But anyway, the thing you are describing around the cards looks like dot crawl.

[ncahammer]

I did try thetoof's suggestion, but without success, the issue remains.
Tried also GuavaComb, Decrawl and DeFlicker after TFM or animeivtc, nothing again

[thetoof]

Hmmm, I didn't go far into testing before posting, so I must've missed the frames you're talking about. Can you tell us the exact frame start/end of the trembling?
and yeah, it's called animeivtc because issues with anime telecining is what made me think of that function a few months ago, but it can also work with live action

[ncahammer]

Hmmm, I didn't go far into testing before posting, so I must've missed the frames you're talking about. Can you tell us the exact frame start/end of the trembling?
and yeah, it's called animeivtc because issues with anime telecining is what made me think of that function a few months ago, but it can also work with live action

Yes, a bit puzzling this name :)

frame numbers are without IVTC/decimation (MPEG2Source only)
Frames 44-59 : The card "9" and some edges trembling/flicker
125-169 : Card's edges trembling/flicker, the "10" number is hardly recognizable in some frames
599-653: the same
780-877: the first case again, the "9" in some frames looks more like "8" or "0"

frame 878: I don't know what is this, blended ?

Thank you very much for your time

[thetoof]

Maybe there are other ways, but anti-aliasing + temporal degraining helped a lot.
The settings for MVAnalyseMulti are a bit crazy... play with it to find a good compromise between speed and quality.
You can lower pel, the temporal radius and don't use nnedi for interpolation, but I recommend that you don't touch the overlap, as it leads to blocking when degraining motion blurs.
mpeg2source("VTS_01_1.d2v")
ivtced=animeivtc(1,aa=2) #the upcoming aa=4 also leads to very good results
nnedipel=ivtced.nnedi(1,true).turnright().nnedi(1,true).turnleft().nnedi(1,true).turnright().nnedi(1,true).turnleft()
mvectors=ivtced.mvanalysemulti(refframes=3,overlap=4,pel=4,pelclip=nnedipel,idx=1)
ivtced.mvdegrainmulti(mvectors,pelclip=nnedipel,idx=1)
Frame 878 is some weird temporal chroma bleeding... it is a bit diminished by mvdegraining.

[Adub]

You might want to look at that code again. b doesn't exist and mvectors is never used.

I assume one takes the place of the other.

[thetoof]

Yup, you are right... I changed the names before posting and forgot to replace that "b".

[ncahammer]

Maybe there are other ways, but anti-aliasing + temporal degraining helped a lot.
The settings for MVAnalyseMulti are a bit crazy... play with it to find a good compromise between speed and quality.
You can lower pel, the temporal radius and don't use nnedi for interpolation, but I recommend that you don't touch the overlap, as it leads to blocking when degraining motion blurs.


That's it! The output is at last watchable!

I will try the options you said to some other scenes that I have the same problem to find a balance

Thank you very much, animeivtc rocks!