Strange DVD pulldown pattern

[yiguro]

Hello,

I have a strange pulldown pattern in a DVD. I noticed it because during static scenes it's easy to spot "vertical jitter", that is, a slight and fast image movement up and down. Stepping through the fields using AviSynth's SeparateFields() in VirtualDub I see this:

a1 a2 a1 b2 c1 c2 c1 d2 e1 e2 e1 f2 ...

Letters are frames. Numbers 1 and 2 mean the top and bottom fields that compose a frame. As you can see, some fields are repeated: the 3rd (same as 1st), the 7th (same as 5th); the 11th (same as 9th)... Those repeated fields take the place of others that are missing, and, if present, would complete a whole, progressive frame: the third should be b1; the 7th d1; the 11th f1... What's going on?

I have cut a sample using VobBlanker and attached it in case you want to see. It is a scene with movement, so it is easy to check the pulldown pattern.

Cheers.

[manono]

That's an odd one, alright. Bobbing followed by decimating works OK:

Yadif(Mode=1)#or QTGMC
TDecimate(Mode=0,Cycle=4,CycleR=2)

But it plays jerky and I think those duplicate fields replaced fields that were originally part of it. Maybe it's an intentional effect, but I doubt it.

[yiguro]

Thanks manono. I tried your script, and it works fine in the sample I uploaded.
Small note: for some reason, I had to add AssumeTFF() in my original VOB. Without it, some frames would play backwards. When I load the sample in DGIndex, it tells me "A field order transition was detected...". Maybe that's why I need AssumeTFF in the original...

Yadif still has some of the "vertical jitter" issues I talked about. QTGMC is slower but mostly gets rid of this problem, and the ouput is very nice. I'm glad you pointed it out in the script's comment.

Anyway, it's a pitty the DVD seems to be badly encoded... This was done by "SMV (Sony Music Video) enterprises" in 2002, so I would have expected some quality control...

Cheers

[manono]

Small note: for some reason, I had to add AssumeTFF() in my original VOB. Without it, some frames would play backwards.

Right, it worked fine without setting the field order in the sample. AviSynth assumes BFF unless told otherwise. Sometimes in the cutting the field order gets reversed. Glad you got that one figured out and glad the video looks as good as it can. The only possible improvement I can think of would be to try and recreate those missing fields/frames by interpolating from what's left, but as you'd be interpolating two fields in a row most likely it wouldn't work so well.