Anything better than FixBlendIVTC for this?

[Lyle_JP]

Okay, so I know the original source is 16mm film. And the final frame-rate is 29.97. But rather than a straightforward telecine... actually, I don't know what the hell happened here. There are some frames intact, some just combed, some blended, and some combed and blended. Oh yeah, and did I mention that it's also macroblock city? So do I deblock first, decomb, then deblend? And what filters to use? I tried using variations of Srestore and Restore24, but I don't want to use a process that bobs, since there are plenty of full-resolution frames that don't need any processing at all, and because both of these filters repeated some frames and dropped others, leading to judder.

FixBlendIVTC gave me the smoothest motion, but the "reconstructed" frames look dreadful. I don't know if its because its old, I'm using it wrong, or if the source's bit-rate is just too low to provide enough detail. So, does anyone have better suggestions? This isn't animation, but should I try AnimeIVTC?

Source (clip): http://www.megaupload.com/?d=U186I551

[nm]

I tried using variations of Srestore and Restore24, but I don't want to use a process that bobs, since there are plenty of full-resolution frames that don't need any processing at all, and because both of these filters repeated some frames and dropped others, leading to judder.
srestore(frate=23.976) worked perfectly on your sample clip.

[manono]

FixBlendIVTC is the wrong thing to use for this as it was never telecined. It's fieldblended from a film source. This works OK:

Yadif(Mode=1,Order=1)#or your favorite bobber
SRestore(Frate=23.976)

If you want to unblend it you have to use something that bobs. If you don't care about the amount of time it takes, then bob with some variation of TempGaussMC. If I'm not mistaken (someone correct me if I'm wrong), the progressive frames get returned intact. I think. In any event, SRestore does one heck of a better job with this than does FixBlendIVTC.
So do I deblock first, decomb, then deblend?
Eh? There's no place in this for a decomb, whatever you might mean by that, unless you mean bobbing. Is the deblocker you use interlace aware? Me, I unblend first followed by any other filtering I might want or need. Usually.

Edit:srestore(frate=23.976) worked perfectly on your sample clip.
Not without a bobber first it doesn't. I guess that's what you meant.

[Didée]

... TempGaussMC ... If I'm not mistaken (someone correct me if I'm wrong), the progressive frames get returned intact. I think.
Here comes your correction. TGMC does not leave progressive frames "intact". Quality shouldn't be bad (hopefully good), but TGMC has no kind of "don't touch this" logic. Everything gets full-fledged processing.

Tdeint has detection for progressive frames, via "tryweave=true". YMMV, since its purely threshold-driven, but the option is there.

[nm]

srestore(frate=23.976) worked perfectly on your sample clip.
Edit:
Not without a bobber first it doesn't. I guess that's what you meant.

Sure. I assumed (perhaps incorrectly) that Lyle_JP already got that part right, but his judder problem was due to using srestore without frate=23.976, which this conversion requires.

[manono]

TGMC does not leave progressive frames "intact". Quality shouldn't be bad (hopefully good), but TGMC has no kind of "don't touch this" logic. Everything gets full-fledged processing.
Reading it over again, I was a bit unclear. When I guessed that the progressive frames remain untouched, I meant the SRestore side of it. So it has no way to recognize progressive frames (or, rather, 2 duplicate unblended bobbed fields side-by-side), and always returns one of the fields instead? Unless you use TDeint as the bobber?

Thanks for the correction.
...but his judder problem was due to using srestore without frate=23.976, which this conversion requires.
Used with a bobber and default settings it gives 25fps with extra dupe frames, but no dropped frames, at least with the provided sample. My guess for the dupes and dropped frames is either filtering before unblending, or incorrect field order.

[Didée]

Srestore has little to do with that, it works "purely progressive" anyway ... hence the bobber needed beforehand. Srestore just "selects" frames from its actual input, so it won't alter any frames if the bobber didn't alter them. (Well, except for the duplicate-averaging modes of Srestore, but that's not really the point.)

[Lyle_JP]

Thanks for everyone's help so far. You know, I did use srestore with the framerate set at 23.976 originally, but I think I ran the video through Decomb without decimating, instead of bobbing.

Which leads me to my stupid question of the day. It was mentioned that there was no telecining of my source, and no decombing should be necessary. Still, there is interlacing galore when I use Honor Pulldown Flags to create the d2v file in DGIndex. That may not be clear from the raw m2v file I provided. So, is Honor Pulldown Flags still the right way to frame-serve this, or is this a rare instance when Ignore Pulldown Flags would serve me better?

[Nightshiver]

Can you post the script you are trying to use?

[Lyle_JP]

:thanks:Okay, thanks to everyone's help, the following code gives me a nice 23.976 fps stream that appears to properly return this video to its film origins without judder or (too much) blending:

TDeint(mode=1, type=2, full=false, tryWeave=true)
Srestore(frate=23.976)

Simple but effective. Apparently, my original problem was insisting on using filters like Decomb and TFM, when that wasn't what was called for. Now I just need a way to get rid of all this god-awful block noise without going jihad on what little detail is underneath.