Need help with Spawn (DVD Season 1)

[Chainmax]

When serving the avs to VDubMod with just the mpeg2source line, I see only 1 interlaced frame every 5 and not 2 (which is what appears in plain 3:2 pulldown, right?). After aplying Telecide and KernelDeint as a postprocessor there is still a lot of combing in some places, although it seems not to happen at periodic intervals. Lastly, I think I saw some ghosting in a couple of frames.
Long story short, I'm stumped. Can anyone suggest me what to do here (especially how to confirm if blending is taking place)?

P.S: I own the DVD (it only costs 9 bucks, after all :))

[gizmotech]

I have seen a vob sample of this cartoon from a friend.

The series contains real interlacing, as well as full frame blending (in spots).

Gizmo.

[Chainmax]

But if it was true interlaced, wouldn't I see pretty much only interlaced frames in moving scenes? Also, what do you mean by "full frame" blending? How can I accurately detect all the places where it takes place and how do I get rid of it?

[gizmotech]

ok, maybe "real interlacing" isn't the correct word, but the source contained a number of areas which were non matchable, and strictly interlaced content. Full frame blending, well that's where you watch through the source, and fix the frames manually. Identifying it is easy. Ususually looks like 1 frame super imposed onto another, commonly a poor IVTC transfer.

Gizmo.

[Chainmax]

So, what you're saying is that I shouldn't use Telecide() at all? While we're at it, what Deinterlacer would you recommend me to use? KernelDeint didn't even notice those interlaced frames I talked about. Should I turn to scripted deinterlacers like SSTomsMoComp or TMCkernelbob?

As for the blending, even though I checked only a portion of the movie, I saw only a couple of frames with ghosting and it was almost unnoticeable. IMO, it's not worth stepping through every frame of a 140min movie in order to repair it.

[Chainmax]

Well?

[Chainmax]

bUmP.

[Wilbert]

Could you upload a part of the vob (30 frames) somewhere?

[Chainmax]

I'm a little short on HDD space now, but I'll try to rip a vob and make a 3sec clip using vobedit soon.

[Chainmax]

Ok, I made three clips. Clips 1 and 2 show the interlacing and are 4.66MB and 4.46MB in size respectively, whereas Clip 3 is where I think I spotted some blending and is 5.19MB in size.

[Chainmax]

Is anyone interested in testing these samples?

[Aktan]

The fields themselves had combing! You can see it using only bob(). This cannot be fixed by any filter.


MPEG2Source("TheSource.d2v")
Bob()


-Aktan

[Chainmax]

There's combing in the fields? That's not a common issue, is it?
There must be something I can do to at least reduce the problem, right?

[DarkNite]

OK, I spent $5.50 at WalMart and got this God forsaken DVD. Now that I have it I've decided the earlier VHS release was most likely the source for this DVD. :devil:

I'm at least positive they did all their editing in video for the "uncut" scenes.

If I reach any satisfactory, non-comical conclusions I'll let you know. Though, at this moment, I'm not sure there is a solution for this source that doesn't involve flame and 9mm bullets.

[Wilbert]

Yes, the ghosting is in both fields. There is not much you can do. You can try the vdub filter Exorcist:

http://republika.pl/voidon/virtualdub/index.html

But I don't know how good it works. So, try it and tell us!

[Chainmax]

Well, I can't try it out myself: my computer freezes up in the middle of all encoding jobs I try to make, even though I just got an extra 256MBs of ram :(.

[Chainmax]

I have an idea: wouldn't it be possible to deinterlace each field separately, weave the result and then IVTC (or whatever else applies)? Would something like SeparateFields().testbob().Weave().whateverelse do the trick?

[Chainmax]

Well?

[Mug Funky]

given that:

a) the chroma combing only occurs on a very small number of frames

and

b) the source quality is horrid

we can simply blur the chroma channel and the combing is gone.


yv12convolution(horizontal="1",vertical="1 2 1",Y=2,U=3,V=3,usemmx=true)


add this after a regular telecine. i've tried this and didn't notice any bad side effects (it doesn't matter overmuch if chroma is blurred).

it looks like this came from DV and an incorrect chroma upsampling DV codec (hellooooo canopus). that must be real fun to watch on a TV.

there's probably better ways to handle this, but with the oddness of the chroma-upsampling bugs out there, it would be very difficult, and honestly the source isn't worth it.

btw.. i wouldn't be so ready to bump a thread. usually people don't reply because they have no idea, or simply can't help. bumping usually doesn't change this. but don't be worried :)

[edit]

hmm... problem disappears when i use my oh-so-dodgy NTSC -> PAL blend-o-riffic script.



odd=doubleweave().selectodd().telecide(order=0,guide=1,hints=true).kerneldeint(order=0,sharp=true,threshold=0)
even=doubleweave().selecteven().telecide(order=1,guide=1,hints=true).kerneldeint(order=1,sharp=true,threshold=0)
interleave(even,odd)

converttoyuy2().convertfps(50).converttoyv12()
unblend()
changefps(25)

[Chainmax]

The problem is that I don't know if I even have to IVTC as I couldn't find a pattern. What will a relatively untrained eye (like mine) notice from this chroma blurring?

hmm... problem disappears when i use my oh-so-dodgy NTSC -> PAL blend-o-riffic script.


code:--------------------------------------------------------------------------------

odd=doubleweave().selectodd().telecide(order=0,guide=1,hints=true).kerneldeint(order=0,sharp=true,threshold=0)
even=doubleweave().selecteven().telecide(order=1,guide=1,hints=true).kerneldeint(order=1,sharp=true,threshold=0)
interleave(even,odd)

converttoyuy2().convertfps(50).converttoyv12()
unblend()
changefps(25)

--------------------------------------------------------------------------------
So, you're saying I should use this instead? Why change the framerate to 25 and not 23.976?

About the bumping, DarkNite said he'd give this a try and post results if he found anything worthwile. That's why I don't want this thread to disappear from page 1.

[Mug Funky]

well, you don't have to change the rate to 25... that's my preference as i like to watch SVCD PAL...

so drop the convertfps line and change "changefps(25)" to "decimate(2).decimate(5)"

decimate slows things a bit, and you _probably_ don't need the full-rate telecide happening, as it slows things a lot too. (i do this because i like how field-blended pans work in anime, but hate field-blended animation, but am too lazy to make a script to distinguish them both, so i simply blend and then unblend... it's roundabout and inelegant but is giving me good results for NTSC->PAL)

it's possible you could simply:

telecide(order=1,guide=1,hints=true).kerneldeint(order=1,sharp=true,threshold=0)
decimate(5)


as the source seems to be film and nothing else (cel animation with horrific telecine)

if you still get the ugly chroma problems, then add the yv12convo i posted earlier after the above. sorted.

[Chainmax]

it's possible you could simply:


code:--------------------------------------------------------------------------------
telecide(order=1,guide=1,hints=true).kerneldeint(order=1,sharp=true,threshold=0)
decimate(5)
--------------------------------------------------------------------------------
I'm not sure that's a good idea. As I said in the first post:
I see only 1 interlaced frame every 5 and not 2 (which is what appears in plain 3:2 pulldown, right?). After aplying Telecide and KernelDeint as a postprocessor there is still a lot of combing in some places, although it seems not to happen at periodic intervals.

So, I don't know if this is actually FILM (24p, right?), VIDEO (30i, right?) or just a telecine (is that 30i as well?).

As I said a couple of posts earlier, wouldn't it be possible to deinterlace each field separately, weave those deinterlaced fields and then IVTC (or whatever else applies)? If so, would something like SeparateFields().testbob().Weave().whateverelse do the trick?

[Chainmax]

Would it?

[Mug Funky]

it's possible, but there aren't all that many frames like that. if it were me doing the backup, i'd just IVTC (apparently SmartDecimate works better than my method. i haven't tried it yet) and do some chroma blurring. the source isn't worth much more than that in my eyes...

of course, do what works best for you :)

maybe kerneldeint the fields... it'll be much faster and should keep all the detail. i'm just not sure it'll detect combing in chroma. a blur would be almost as good and much less fiddly.

[Chainmax]

Ok, I'm going to try the following as soon as I get a chance:


SmartDecimate(24,60,bob=DGBob(mode=0))
yv12convolution(horizontal="1",vertical="1 2 1",Y=2,U=3,V=3,usemmx=true)
HybridFuPP(640,480,preset="high")

Thanks for your patience and all the useful input:)