weird decomb behavior

[lipton]

I'm trying to IVTC a HK dvd and there are several scenes that decomb doesn't want to deinterlace (this is an anime DVD). I'm using mpeg2dec to stream the d2v file. Here is the decomb line
Telecide(agg=true,chroma=true,debug=true).Decimate(mode=2,cycle=5,show=true)
I've tried reversing the telecide fields and lowering the threshold but it won't pick up the combing

Here (http://booboo.netfirms.com/decomb_bug.zip) is the decimate and telecide debug output.

The error occurs at frame 3418 (all the frames after that for about 2.5 seconds are detected as clean)

Good frame

http://booboo.netfirms.com/DC_ok01.jpg

Bad Frame (frame right after good frame)

http://booboo.netfirms.com/DC_bad01.jpg

Bad Frame (frame right after first bad frame)

http://booboo.netfirms.com/DC_bad02.jpg

Edit: added 1 new picture that decomb picked up as new instead of dup
Edit2: THe problem is with the dvd source itself, looks like a really bad mastering, kinda hard to image a HK DVD would have that :p It looks like a shift in one of the interlacing fields and decomb doesn't pick it up :(
Feel free to delete this thread if you wish moderator

[Guest]

Let's get the field matching right first and then worry about the decimation. You specified agg=true but you need to set guide=1 also or agg has no effect. Run this and repost the log for that area:

Telecide(guide=1,gthresh=30,chroma=true,debug=true)

Just leave Decimate off for now until we get the matching right. Let's try with just this and then we can add agg and mm if needed.

By the way, it did deinterlace those frames. But it didn't match them correctly.

[lipton]

Here (http://booboo.netfirms.com/decomb_bug2.zip) is the new output. I ran it 10 frames before upto a few frames after it ended.

[Guest]

OK, it's too hard to figure stuff out like this. Please make a source clip or VOB available for download. Thank you.

[Guest]

OK, I had a look at your clip. That clip is so crappy there is nothing you can really do to make it look good. First, it does not show any 3:2 pattern, so you can't use guide=1. Second, it has the dreaded blended fields problem, which means you must deinterlace with blend=true. But third and worst is this: Use separatefields() to view the fields. At frame 34 (after separation), as you single-step through the fields you will see a weird oscillating spatial distortion set in. This is not just simple field bobbing, it is some horrid artifact, probably from capture. I have seen it before.

There is nothing Decomb (or any other software) can do to improve this. Sorry.

Just out of curiosity, what is the history of this clip?

[lipton]

The source is a Hong Kong DVD from a series called "Detective Conan". Clip I sent you was from the first episode (timeframe of the problems was 1:53 - 1:57 into the episode). The source clip I send you was unaltered (mpeg2source of the .d2v encoded with fast-recompress)

[scmccarthy]

@neuron2There is nothing Decomb (or any other software) can do to improve this. Sorry. VerticalReduceBy2 would still work, would it not? Or does that oscillating effect even mess that up?

Stephen

[Guest]

No that will not help this clip. You wanna see it?

[Kaiousama]

To have a realistic view of the incidence of this telecine problem proposed by Lipton, you have to know that almost any HK anime dvd has one or two of these screwed up scenes for every episode (in a hk dvd normally are stored 6-9 episodes).
Actually (as you easily see) noone of the ivtc filters can solve it, so a decomb that can handle these scenes would be like a miracle on earth ;).
I've tried to analize the telecine pattern but is very strange... it seems that (for a reason i doesn't know) that a field is shifted down (btw.. all these dvd's are bff, i don't remember wich one was shifted) for a limitate time period, often these scenes doesn't begins at scene changes but during a still or a moving image, their duration can be between a few punch of frames and 4-5 seconds of video.

And there are sometimes a even more painful implementation of this bug... praticcally the shifting event appears not in all the frame but it begins in a part of frame and, a little lately in time is extended to all frame (and obviously it make the same when the frames return normal).... it's a very annoying and courious behaviour (i'd like to know what mpeg2 conversion program can reproduce this effect ^_^)

I don't know if these informations could be useful... take it easy ;)

[Guest]

Right Kaiousama. It is really bad. In the clip I saw, it is as if in one field the top half is vertically compressed and the bottom half expanded, and then in the next field the bottom is compressed and the top expanded. It's just totally bizarre and no software can be expected to deal properly with it.