Nasty Interlacing?

[Dogway]

Hello. Im having problems deinterlacing this source. Im not sure how to deal with it so if someone could point me in the right direction. I cant explain the synthoms, it just looks like really nasty stuff. no pattern, intra-soft-blending... Dont know how to call it.

Right now Im trying with TFM, but can I deinterlace it with TMPGMC? I have read all over the web its the best one out there. I noticed some kind of blending, so I also tried srestore, but no success...

Sorry, but when it comes to deinterlace, Im a total newbie...

http://www.sendspace.com/file/alnaxz
http://www.sendspace.com/file/yfauxg


So far this is all I have tried:


#AssumeFPS(30000,1001)

#TempGaussMC_beta1mod(tr2=3,mt=true)
#selecteven()
TFM(order=1,slow=2)
#Yadif(Order=1,Mode=1)-
#vinverse()-
#SRestore(FRate=24)-
#SRestore(omode="""pp3.Vinverse().Clean().MDegrain()""")
#tdeint(hints=false,type=1,slow=2,emask=tmm(length=6,ttype=0,mtqC=0,mthC=0))
#TDecimate()

[manono]

First, use anything besides Megaupload. Try Sendspace or Mediafire next time. Second, no one wants to download 318 MB when probably 25 MB is more than enough to tell what's going on. Open the VOB in DGIndex, scroll to a place with motion and use the [ and ] buttons to mark off a 10 second or so section. Then File->Save Project and Demux Video. Upload the resulting M2V.

[Dogway]

Thank you, I will try that.
then, what's the problem with megaupload? :s

[manono]

What's the problem? First you have to type in numbers and then you have to wait 45 seconds. No thanks.

[Dogway]

XD I see... Actually it has other good points: speed, visual clarity, huge storage. I will post new link by tomorrow

[manono]

And did you test your link before posting? Even if I wanted to download it (which I don't), I couldn't. I can't anyway.

If you want people to help, make it easy for them.

[Dogway]

Updated: http://www.sendspace.com/file/alnaxz

The link in megaupload was working, around 10 people downloaded it. I just upload where I'd like to download from, sorry for not share the forum likes, I've been here only for a few months.

[manono]

The link in megaupload was working, around 10 people downloaded it.
So, where are the responses from those 10 people? If it's counting the number of people that tried to download it, then 5 or 6 of those attempts are mine.

There's not enough steady movement for me to be positive, but I think it's field-blended from a film source. While no unblender is perfect, this does a reasonable job:

Yadif(Order=0,Mode=1)#Sample is BFF.
SRestore(FRate=23.976)

http://avisynth.org/mediawiki/Srestore
http://forum.doom9.org/showthread.php?t=95924

[Dogway]

Yes, thats almost as near as I could get, but still there was some ghosting/blending. I guess that's the most I can get with it...
Cant I use TFM? Yadif seems a bit unstable on my system...

TFM(Order=0,Mode=1,Slow=2) ?


I think this sample is better:
8 Mb
http://www.sendspace.com/file/yfauxg

[Dogway]

Fixed the code a bit:

Removed further ghosting
Removed residual aliasing
Removed Cplugin Yadif need
Order seemed to be TFF...

edi=NNEDI2(field=-2) #antialias
YADIFMod(order=1,mode=1, edeint=edi) #deinterlacing
SRestore(mode=-4) #blend-decimation
TDecimate(rate=24,mode=2,maxndl=30) #dup-decimation
Dup(threshold=15,blend=false,copy=true,chroma=false,blksize=8,show=true,debug=true) #dup-stabilization: replaces first frames with seconds, normally the ones showing blend.

using mode=-4 discarded some annoying blending, but allowed other shallow blending being introduced. Actually, using only frate or mode just works as a dup filter, isnt it? Which takes me to the question, what is better for decimation? Srestore or TDecimate?

Here you can see a test: 1Mb http://www.sendspace.com/file/ebc9zc

[manono]

Thanks for the second sample. It's much better and now it's easy to tell that the framerate is supposed to be 23.976fps.
Cant I use TFM?
No. TFM is a field matcher and there are no fields to match. It can't be IVTC'd, but only unblended. Thank the idiots that created the DVD for that.
...what is better for decimation? Srestore or TDecimate?
You should definitely do the whole thing within SRestore. I think it decimates using TDecimate anyway. And the framerate's 23.976fps, not 24fps, not that there's much difference. SRestore works better by setting the correct framerate.
Order seemed to be TFF...
Although the entire video may very well be TFF, both samples were definitely BFF.

[Dogway]

I will stick to what you say. Anyways yadif gived me aliasing issues, which I tried to fix with nnedi being called by yadifmod. Now, yadifmod doesnt like BFF, it keeps fields interlaced(in preview), it crashes on encoding.

edi=NNEDI2(field=-2)
YADIFMod(order=0,mode=1, edeint=edi)
SRestore(frate=23.976)

The test of the first code you gave me:

Yadif(Order=0,Mode=1)
SRestore(FRate=23.976)

This is the test: 1Mb http://www.sendspace.com/file/380li1

It might be BFF but it has lots of artifacts compared to my version. Although mine has denoiser, color correction, etc, regarding field-blending it doesnt look quite nice.

[manono]

The field order is wrong in that MP4. I can only go by your samples, which were BFF. If the whole video is TFF, that's what you use. The field order probably got reversed when you cut the samples. You know how to determine the field order, don't you?

And I agree that a better bobber should be used with that one. I use Yadif for testing because it's fast.

[Dogway]

Normally I just test with TFF, if something goes wrong, I try BFF. I read about detecting field order long ago, since what I can remember in BFF you see your fields jittering.

What better deinterlacer do you suggest for this, might I say bobber? I can only come up with yadif, tfm, mcbob, tempgaussmc, and tdeint. maybe tdeint? is yadifmod better?

I will try to use a better deinterlacer with BFF and see what happens, I will ensure it is BFF by the way.

Also, DgIndex warned me about this: Field order transition: 0 -> 2
Is it relevant? Im using this "fixed" .d2v in my tests.

[manono]

YadifMod is better. TempGaussMC might be best.
I will ensure it is BFF by the way.
If that's what you used last time, it was wrong. Use the correct one.
Is it relevant? Im using this "fixed" .d2v in my tests.
It fixed it so it was the same all the way through. Apparently they switched somewhere along the line.

[Didée]

Just to note - if you find MegaUpload annoying, have you ever been faced with RapidShare? :D

Infact I was one of the downloaders of that sample (no problems) - but when I saw it's animation, my interest immediately disappeared ... (it's not my thing to *guess* for a possible pattern, when you can't be *sure*, at least not easily, because of all the dup frames that are inherent to animation. Leave that task up to the Anime enthusiasts.)

[manono]

Just to note - if you find MegaUpload annoying, have you ever been faced with RapidShare? :D
Hehe, you'll notice that I didn't mention Rapidshare as a possible alternative.

Really, you were able to get the original sample? All 318 MB of it? I must have clicked on it half a dozen times and never got anywhere. Not that I was at all interested in getting the whole thing, but was curious as to why I couldn't download it.

Here's a comparison of some of the file hosting sites. The red box under 'Direct Download' steers me away from those places:

http://en.wikipedia.org/wiki/Comparison_of_one-click_hosters#Comparison_of_notable_file_hosting_services

I got a free account at MediaFire partly because of the direct downloads it allows others, partly because of the unlimited amount of storage it provides.

[Didée]

Yes, the DL from MegaUL went without any problem. Also, it was rather fast ... like 500~600 kB/s, IIRC. Mediafire usually gives me half of that, on the better days.

[Dogway]

Fixed the code a bit:


Well I will stick to that code, (if nobody suggests me something else), with modifications: using 23.976fps instead of 24. Also dropped the srestore(mode=-4), and tdecimate, it was causing some dups. So finally I just keep the Dup filter for some blends artefacts and whatnot.

edi=NNEDI2(field=-2,qual=3)
YADIFMod(mode=1, edeint=edi)
SRestore(23.976)
Dup(threshold=9,blend=false,copy=true,chroma=false,blksize=8,show=false,debug=false)

Definitely its TFF, or yadifmod(order=1) if that makes sense. Although checking the source it has the BFF jitter I talked before.

Thanks all the way.