Decomb (cartoon / missing fields)

[WarpEnterprises]

I have the following problem case with a capture (telecined PAL, numbers denoting the restored frames, A/B the composing fields)
1a, 1b, 2a, 2b, 3a, 4a, 4b

So field 3b is missing and Telecide picks (depending on the settings) 2b or 4a as partner field which is of course both wrong, the matching value (show=true) is correctly detected as quite high (10000 instead 500 else)

Is there a way to use this field as the whole frame (bob-deinterlace it) or to react in another way if the matching value is too high for a true match?

Or is this not possible as the final framecount would depend on the number of missing fields?

(The sad thing is that the missing fields are not always top or bottom, so reverse does not help)

[Guest]

I am having conceptual problems with some of your assertions. :)

To cut through those can you please post a link to a part of the clip (unprocessed source of course) that shows the problem. Then I can properly advise you.

Do you have postprocessing enabled? If so, doesn't that fix up the frame?

[WarpEnterprises]

I hope a PicVideo AVI is ok (huff is too big for 56k).
The problem originates I think from a 24/25 fps conversion.
I would like not to use postprocessing because the video looks like it effectively has only 288 lines, so ONE field would already be at the max. resolution (so I decided to use only VCD target resolution)

I also tried

SeperateFields.Bob.Decimate(cycle=2, mode=2)

but this too throws away the single field.


AVI (http://members.aon.at/archi/wepictures/winnie.avi)

thx

[Guest]

Your clip looks ike this:


ABCDEFGHIJKLMNOPQRSTVWX
ABCEFGHIJKLMNOPQRSTUVWX
...1...............2...At point 2, one of the D fields is deleted. But Telecide is OK because he starts matching to next. At point 2, one of the U fields is missing. Telecide cannot find a match because he matches only on the bottom field (a six-way search would allow Telecide to output a complete T frame). Normally, postprocessing would automatically handle this. For example, I set post=true and threshold=30 and only this frame was deinterlaced. So I don't understand fully why you want postprocessing off, since with a good threshold you'll only touch the bad cuts.

Even if you do want postprocessing off, you can use a manual override to force just this frame to be deinterlaced (see help file). If you use blend=false that will give you the effect you want ("use this field as the whole frame"). The downside is that it is not automatic. For an automatic solution use postprocessing with an appropriate threshold as described above!

One further interesting point. If you decide that a manual solution is OK because there are not too many cuts like this, it is better to use FreezeFrame() because then you will keep full frame resolution. E.g.:

Telecide(post=false)
FreezeFrame(18,19,18)

It is interesting to contemplate having Telecide automatically repeat the last frame in such cases if it is a clean frame.

Or you can just use Avisynth to delete the bad frames.

Of course, no matter what you do, you're going to get a jerk at the bad cuts.

[Guest]

@neuron2

That's all very interesting but what the original poster apparently wants to do at point 2 is make a frame interpolated from field U! Then the clip would not have a jerk there. Telecide always interpolates on a given field and if that isn't the one you need...:confused:

A similar argument can be raised against point 1.

This is an interesting but difficult problem. I am going to give it some thought. It seems that we need a 6-way match followed by interpolation on the orphan field for frames still combed.

[Guest]

OK, I had a little think during my swimming workout. The best thing we could do with this clip is recreate and reinsert the missing D and U fields. This would certainly be possible. You'd have to slow down the audio with an external program of course.

Obviously Telecide can't do this because it doesn't add frames. Before I contemplate making a new filter for this, please tell me if there is a regular pattern to the deleted fields. For example, if this is resulting from a 25fps to 24fps conversion, you'd expect to have two fields per 50 deleted. If there is a regular pattern we'd know the right frame count to set. Otherwise we have to solve that differently.

One practice seen more these days is deletion of fields to effectively shorten a program and thus allow more room for commercials. I wonder if this is going on here.

So please advise about the deletion patterns you see.

[WarpEnterprises]

Yes, now we mean the same. :)
Sadly the deletion is often with non-trivial patterns.
It is difficult to count because the foreground action is often at ~12fps.
But I saw those missing fields each 30-24-30-24 or 33-31-33-33-33 and other patterns.

Could you think of a way to use a 2pass mechanism:
1st: detect (bobbed) fields without duplicate (e.g. write the result of Dup to a file)
2nd: fill in the fields considering the result (and adapt audio, yes)

(there are blended fields too in this video but thats another story)

And No, this must have other reasons that shortening the clips.

[Guest]

Originally posted by WarpEnterprises
(there are blended fields too in this video but thats another story) It's part of the same story because the blended fields are going to make it near impossible to reliably detect the missing fields, because the effect of a blend and the effect of a missing field will both cause high match values for all three match attempts.

How frequent is material like this?

[WarpEnterprises]

Ok, I see.
Can't tell you statistics for now.

I just figured out how to use debugview & Dup, so I will try if I can figure out a pattern and a possible solution.

[Guest]

Originally posted by WarpEnterprises
Ok, I see.
Can't tell you statistics for now.

I just figured out how to use debugview & Dup, so I will try if I can figure out a pattern and a possible solution. I meant how often do clips like this occur in the real world. If it is often enough it might be worth trying to make a filter. But if it is just a single oddball...

[vidiot]

Hi, I´m just curious:

- Is the winnie pooh clip fron digital tv?
- Are you able to detect the jerkiness on tv? Or just on the computer?

-----------
I captured nearly the hole series from analog tv for my kids and
the telecide filter for VD was very helpful then...

Regarding to the clip:
I did try PAL Frame Restorer from Simon Walters - and it worked as good as telecide (the "old" VD one).

Happy eastern!
Harald

[WarpEnterprises]

@neuron: at the moment I don't kown if this happens often, but I will tell if I encounter it more often.

yes, its DVB. You can't see the jitter on TV because one FIELD ís present, but if you try to make the telecined video progressive (even using only 1 field) these troubles arise.
But I've found a solution without too much manual typing:
1) I let Dup run over a SeperateField.Bob'ed video and create with DebugView a file with the matching values
2) I import this to excel and run a macro which looks for those single field. These I Duplicate and Delete some fields which are 4times there (too stay in sync)

Now I have a perfect VCD-resolution video (when WarpSharped it looks really good enough).