PAL > NSTC Problem.

[n0an]

So while studying PAL > NTSC I stumbled upon a source which in a cycle of 6 has 1 progressive + 5 Interlaced frames. I recalled manono's replies here (http://forum.doom9.org/showthread.php?t=151698) and followed the instructions.

[sample]
http://www.sendspace.com/file/2z0sx7

The following is what I have tried so far and I am now totally confused which works best for this.

DGDecode_mpeg2source("C:\Users\Admin\Desktop\DVD\VTS_02_4.d2v")
a=Yadif(mode=1).Srestore(frate=24.975) #Yadif(mode=1).RePAL()
b=yadifmod(field=1, order=1, mode=1, edeint=nnedi2(field=-2)).Srestore(frate=24.975)
interleave(a,b)


Yadif and Yadifmod don't show much difference, but they do leave some residual interlacing. Vinverse() helps in that.

[Blue_MiSfit]

I looked at this source, and the usual bob +repal / srestore didnt seem to fix things. there is a lot of residual aliasing after deblending. I can't even get a bob to look clean (ignoring blends of course).

I think some kind of field matching needs to happen first? I dunno lol I'm not much of an expert in this stuff..

Hopefully some clever soul knows the source.

[manono]

What's wrong with a simple:

Yadif(Order=1,Mode=1)
RePAL()

or :

Yadif(Order=1,Mode=1)
SRestore(FRate=24.975)

Or the same using your YadifMod script? I see neither any remaining interlacing nor much remaining blending. Aliasing is a different matter. You can thank the lousy Indian source DVD for that. After all these years they still haven't learned how to put out even a mediocre DVD, much less a good one.

Please choose your samples more carefully. Photos on a computer monitor, people sitting in front of computer monitors, and people talking on the phone, don't really provide the movement necessary to help figure out what's going on. Only about 10% of that 50 MB sample was at all useful.

[n0an]

My cousin brother suggested me this from meGUI and then the fps from the PAL > NTSC idea.

tfm(order=1).tdecimate(hybrid=3)
Srestore(24.975)

What do you think this looks good. What do you think?

Will be careful about the sample.

[manono]

What do you think?
At the risk of offending your cousin brother, that script makes no sense at all. I ask again, what's the matter with one of the basic PAL2NTSC unblending scripts?

[n0an]

At the risk of offending your cousin brother, that script makes no sense at all. I ask again, what's the matter with one of the basic PAL2NTSC unblending scripts?

It's not about offending anyone, but about what is right. That suggestion was given by meGUI (my cousin did it on his pc), and Yadif().tdecimate(hybrid=3) gave 30fps. Then Srestore(frate=24.975) was used to restore to the proper fps.

Hybrid 3 = blends conversion of 24p sections into 30p and leave 30p untouched. Why is this not making sense?

[manono]

I know what Hybrid=3 does. The result of that script is a blended, blurry, and jerky-playing mess. You don't use TFM because there's nothing to field match. About all it's doing is deinterlacing it with TDeint. And then after that The TDecimate line doesn't do much, if anything at all, except maybe create even more blending. And then after it's all been made progressive, there's nothing left for SRestore to unblend, so all it can do is pull out frames.

Following a MeGUI script is about the worst thing you can do with this kind of source.

[Nightshiver]

n0an, how about following the advice of those with more knowledge than you, hmm? If you want some more concrete answers, provide us with a sample.

[manono]

He's got a sample linked in the first post. Just using the regular stuff I didn't see much wrong except for the aliasing Blue_MiSfit mentioned. Maybe there are difficult parts elsewhere in the movie.

[Nightshiver]

Ah, didn't see that it was a coded link.

[Didée]

Aliasing is a different matter. You can thank the lousy Indian source DVD for that.
Looking at the sample, I have to disagree. Apart from the dreaded blend-conversion, the source actually seems quite okay. In fact, the aliasing is a problem introduced by the bobbing-filter/s, due to the rather good sharpness of the original footage.


In contradiction to some of my own past advises (*cough*), try the following:

mpeg2source("VTS_02_5.d2v")
o = crop(0,56,-0,-56)
tgmc = o.tempgaussmc_beta1u(1,1,1,edimode="nnedi2",blocksize=16,sharpness=1.33,SVthin=1.25,truemotion=true)
bob1 = o.bob(0,0.5)
bob2 = bob1.clense(reduceflicker=false).merge(bob1,0.5)
bob3 = bob2.mt_adddiff(mt_makediff(bob2,bob2.clense(reduceflicker=false),U=3,V=3),U=3,V=3)

tgmc.Srestore(frate=25.0,dclip=bob3) # edit: 1st clip arument (tgmc) was missing ...
... what aliasing?

[Didée]

The old Abakus has finished ... result of the script proposed above: http://www.mediafire.com/?ym5zgymzgm2

Not completely free of errors, but then, blend-restoring hardly ever is. Overall, that seems quite acceptable to me: fluid appearance, and there's no aliasing or blending left that's apparent on normal playback.

[manono]

Looking at the sample, I have to disagree. Apart from the dreaded blend-conversion, the source actually seems quite okay.
The aliasing is, of course, introduced by the unblending. I was more or less agreeing with Blue_MiSfit's assessment when I made the statement.
there is a lot of residual aliasing after deblending.
And since there's no interlacing left after the unblending, I was also thinking that n0an was mistaking the aliasing for interlacing.
Yadif and Yadifmod don't show much difference, but they do leave some residual interlacing. Vinverse() helps in that.
Ordinarily I wouldn't have even mentioned it because that's the price one pays when unblending these crap PAL2NTSC sources.

So now it's OK to use TempGauss_MCBeta1? :) Maybe the masochists among us will note your change of heart but I, for one, am not willing to wait a couple of days for an encode to complete on the typical 2.5-3 hour Indian movie length. Don't get me wrong; for something short like a music video nothing beats it, but otherwise it's impractical. Just my opinion. Maybe it's time for my computer upgrade.

[Didée]

Maybe it's time for my computer upgrade.
If you do, send me your current rig. For me, that will be an upgrade already.

The aliasing is, of course, introduced by the unblending.
Sorry for being pedantic, but no, not by the unblending. If there's aliasing introduced, then it is due to the bob-deint. Srestore basically is just a decimator, using a pre-prepared bob stream.

note your change of heart
Hehe, no, no change of heart. The fact that mocomp-deinterlacers are badly suited for fieldblended content still holds on - that's sort of a technical necessity. But sometimes you've to trade one burden for another. I'd never recommend to clean a window with sandpaper - but in some hard cases, it might be a suitable method nonetheless. ;)

Also, the topic "speed of TGMC" is well known, sure. The point is that there was so much crying "I get bad aliasing whatever I do" in this thread. So I just showed a possible way to get a smooth non-aliased result, fluid, and with good detail.
In comparison:
- TDeint+NNEDI2 is a low compromise - it's not exactly a sprinter either, and will suffer from remaining flicker on prominent detail (hard threshold misery). (Yadf+NNEDI2 is little faster, but probably not better quality.)
- TDeint+NNEDI2+TMM - probably better in regard to detail stability, but the speed starts drifting toward TGMC range, and it still will have issues.

Yes, field blending is evil. And yes, the "rather safe" methods range from slow to very-slow. But we won't change the fact that Indian movies are like they are. And despite everything, there are lots of people who want to process them. For those, it might be better to have a slow script running for some days without user intervention, instead of personally spending several days at the PC, searching for a fast solution, but to no avail. :)

[Blue_MiSfit]

Excellent reply, Didee! Thanks for your contribution! I know n0an will be happy!

~MiSfit

[EuropeanMan]

Looking at the sample, I have to disagree. Apart from the dreaded blend-conversion, the source actually seems quite okay. In fact, the aliasing is a problem introduced by the bobbing-filter/s, due to the rather good sharpness of the original footage.


In contradiction to some of my own past advises (*cough*), try the following:

mpeg2source("VTS_02_5.d2v")
o = crop(0,56,-0,-56)
tgmc = o.tempgaussmc_beta1u(1,1,1,edimode="nnedi2",blocksize=16,sharpness=1.33,SVthin=1.25,truemotion=true)
bob1 = o.bob(0,0.5)
bob2 = bob1.clense(reduceflicker=false).merge(bob1,0.5)
bob3 = bob2.mt_adddiff(mt_makediff(bob2,bob2.clense(reduceflicker=false),U=3,V=3),U=3,V=3)

Srestore(frate=25.0,dclip=bob3)
... what aliasing?

Hi Didee,

I downloaded this sample...and for learning purposes, I used your script here


DGDecode_mpeg2source("C:\Users\3 Idiots\Desktop\VTS_02_5.d2v", info=3)
Import("C:\Program Files\AviSynth 2.5\plugins\srestore.avs")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\mvtools2.dll")

o = crop(0,56,-0,-56)
tgmc = o.tempgaussmc_beta1u(1,1,1,edimode="nnedi2",blocksize=16,sharpness=1.33,SVthin=1.25,truemotion=true)
bob1 = o.bob(0,0.5)
bob2 = bob1.clense(reduceflicker=false).merge(bob1,0.5)
bob3 = bob2.mt_adddiff(mt_makediff(bob2,bob2.clense(reduceflicker=false),U=3,V=3),U=3,V=3)

Srestore(frate=25.0,dclip=bob3)

crop(0,56,0,-56)
spline36resize(720,304)


I get ugly interlacing lines still? I don't get it...

[thetoof]

After a quick glance, it looks like what is processed by srestore is the original clip, not the bobbed one.

So, replace the srestore line by

tgmc.Srestore(frate=25.0,dclip=bob3)

[EuropeanMan]

^ Well, DUH...I'm a dummy. Why didn't I catch that???!!! Thanks toof :)

[thetoof]

The question I've been asking myself since I saw Didée's post is why didn't he catch that? :p

[Didée]

There is a like 99% chance that I b0rk up a script when doing code re-ordering in the post editor box. But yes, why didn't I catch that? Some questions remain without answer....

[EuropeanMan]

^ If possible, can you guys point me in the right direction re: aliasing. Manono pointed out aliasing with the RePal/yadif solution. I suppose I need to know what is it that I am looking for, and where in Didee's script it is taken care of and how. :)

Will have to do a comparison now with Didee's script vs. YadifMod/SRestore @ 25fps.

PS - thread title is wrong. Should be deinterlacing help. This has NOTHING to do with PAL --> NTSC.