What field matcher to use?

[ministrix]

Hello!
I got a problem ivtc'ing a movie. Its a asian dvd and using either decomb or tivtc do some good things and some bad things. If i do the usual Telecide(order=1,guide=1).Decimate() it looks ok except for some few times during the film when i get some interlacing effects. ( i think maybe the patterns are imperfect and therefore the telecide cant match right) and if i use TFM(pp=7).TDecimate().vinverse() this removes those faulty frames but i get this seesaw effect a few times during the film, chk this pic how the edge of the bridge should really be straight or the right side of the road
http://tinypic.info/viewer.php?file=6yhxz2klapgd5bl7jwlm.jpg
this is how it looks with decomb:
http://tinypic.info/viewer.php?file=0ni2p5orv92elxsq19z6.jpg

So anyone got any ideas of what i can do?

[Blue_MiSfit]

Try the decomb version with Vinverse() as a post processor

[thetoof]

Dunno if this'll help, but it did for hard telecined anime ivtc:

tfm(clip2=tdeint(edeint=nnedi(),emask=tmm(0))

This solution is still "in development", so you can follow its evolution here (http://forum.doom9.org/showthread.php?t=137240). We started talking about it in post #46 & 54

[ministrix]

vinverse() didnt help those interlaced artifacts:( and im digging into thetoof solution but need to find plugin for tmm. Its late so ill dig deeper tomorrow

[thetoof]

It can be found in the TDeint and TIVTC thread (http://forum.doom9.org/showthread.php?t=82264)

[ministrix]

thetoof: actually this looks to wrk.
http://tinypic.info/viewer.php?file=pgy9m6ac8ld3gggsp53n.jpg
much better:)
of course with my old non dual processor i have to encode the whole movie to chk it properly because frame stepping ahead is sloooow. hehe.
But ill report back cause this looks good for non anime as well

[ministrix]

very weird. the preview of the script looked good in vdubmod but when i encoded it it had just the same seesaw problem. what does this?
code:
tfm(clip2=tdeint(edeint=nnedi(),emask=tmm(0)))
TDecimate()
vinverse()

[thetoof]

it had just the same seesaw problem. what does this?
I don't understand what you mean...

If you could post an unprocessed sample of a portion of your clip that is still interlaced after IVTC, it'd be very helpful.

[jeffy]

@thetoof: not sure, stair-stepping?

[ministrix]

@thethoof. Jeffy is pointing in right direction. Its that stair-stepping/seesaw line on railing on left side of bridge. its suppose to be straight. In preview in vdub with ur settings in script, it looked good (chk out the picture in my post nr 6) but however, after encoding it looked like this
http://tinypic.info/viewer.php?file=n7nwlyqlfowopu0bmt2p.jpg
Same problem back again.
Heres a vobsample
http://www.megaupload.com/?d=WMC0B9B4

[thetoof]

This seemed to help. Not perfect, but still...
If you see another "very ugly frame", give me its exact # (or upload another test clip if it's not in the one I have) and I'll try some other stuff.
You can also consider using an anti-aliasing filter.
tfm(clip2=tdeint(2,edeint=nnedi(-2),emask=tmm(1)),slow=2)

[ministrix]

Dude this looks great, right on par with decomb plus the random interlaced frames is removed. Maybe these settings can b used instead of:
TFM(pp=7)
TDecimate()
vinverse()
which ive been using for this kind of material before (but then again ive never seen this stair stepped lines before either). Thanks for the help, I'm not quite sure what nnedi and tmm does but it sure wrks.

[thetoof]

- nnedi doubles the framerate by interpolating every field to create a frame out of them
- tMM creates a Motion Mask at framerate*2
- Tdeint takes the ExernalDEINTerlaced clip of nnedi and the ExternalMASK of tmm as a reference to do his "mode 2 operation": smartbobbed field-matching (same rate output, blend frames from bobbed stream)
- TFM does the field matching with his slowest (and most accurate) function by using the clip Tdeint outputs to do this:
When using PP > 1 this parameter may be used to specify a clip to take deinterlaced frames from instead of using one of TFM's inbuilt deinterlacing methods. If PP < 5 then frames will be taken as is from clip2, if PP >= 5 (i.e. using motion adaptation) then TFM will build the mask as usual and only pixels in moving areas will be taken from the frames in clip2