PlatB

Molly said:

I use QTGMC(“Fast”).SRestore(23.976) for PAL-to-NTSC deconversions, for what it’s worth.

I have never used srestore but I’ll take a look at some point. Repal has been my go-to since development (I helped debug it back then on doom9 under a different handle) as it gets cadence right. There was an alternate fork c.2008 that takes the clip straight from 29.97->25, but in my experience it introduces more obvious stutter (I tried both with the superman clip). The best approach is to get the clip to 24.975 and then apply Assumefps(23.976).

QTGMC seems to be regarded as the best of the smart bob filters, so I expect using this instead of yadif could yield the biggest improvement to the final output.

Recently I found myself working with a NTSC DVD of an eastern European film that had the oddest transfer chain. It was from a PAL source but was standards converted to 23.976 unlike the typical 29.97 (ie the Superman clip, ect). It was a recent DVD master so the 3:2 pulldown flags were 100% accurate without interlacing… but the result was that every single output FRAME (not field) was blended. Even though this transfer method might be cleaner for playback on a 24p capable TV, the side effect is nothing can be done to restore the original (the proper playback speed would be 23.016). Whoever did the mastering clearly had the tools, so why they didn’t slow the 25 back to 24 instead of blending needlessly (what authoring software promotes that?!) is beyond me.

Booshman said:

I’m going with the scaled down option to preserve the 4:3 ratio and so I don’t have to alter the pan and scan composition to account for the reframing.

Yeah, it looks fine. I would be much more concerned about slight resizing BDs of digital films from recent years where pixel-level sharpness is already dialed in.

I managed to re-NTSC the clip using avisynth.
https://mega.nz/#!SwYjiBDD!WcUZxy7NFgKp7se_kubOnXkHtQ40L98cuWErgg8zD1I

mpeg2source(“ABC RIC 2.demuxed.d2v”)
yadif(mode=3) #smartbob
repal() #returns a 24.975 clip
AssumeFPS(24000,1001) #slows the clip back to its original film speed
BilinearResize(640,480, 10.0, 0.0, 704.0, 480.0)

The frame blending is baked in, but the combination of a smartbob (either yadif mode 1 or 3, and they seemed about the same in output) and the old repal plugin is probably as good as it gets.

Booshman said:

Cheers for that, I’ll have a look.

I’ll go through processing on my editing computer tomorrow, but by the looks of it almost certainly possible to use the classic repal plugin for Avisynth paired with a smartbob filter to force out a clean pseudo-PAL 24.975 from 29.97, then slow it back to 23.976 (assuming the BD to match is 23.976) without any temporal anomalies. Since resolution is already OTA VHS, I’d be more concerned about preserving cadence in motion.

For 720p a 92% resize is not going to be a real problem, but it might be worth trying against the crop approach just to compare results for yourself. There are plenty of downsizing filters with sharpening options to approximate the look of the original footage, so I wouldn’t get too hung up about it.

Molly said:

Got a fairly precise tag on that ABC recording: WXYZ (Detroit, MI), April 20 1986; broadcast started at 8 pm, but about an hour is missing, not including commercials.

Man that is something… I have the same Detroit broadcast recorded myself and complete (in LP mode IIRC), but the tape probably got pitched in the last few years since my parents started clearing out the house. Being that it was obviously different from any commercial video release I watched it quite a bit back in the day.

Booshman said:

Does anyone have any tips on upscaling the DVD footage from the RIC? I have magicbullet for after effects, but if there is a better option I’d like to know. I need to increase the the size by 28%, and it would be nice for it to look as good as possible.

I am not sure about magicbullet’s resizing filters, but it might be worth trying against some of the avisynth plugins (a comparison was done a few years ago, archived here: http://archive.is/7bvQj ). One thing to keep in mind is that cleaner ratios scale better ie 25% upscaled probably comes out sharper than 28% (3 lines into every 4) - sometimes a slight black windowbox is the least worst option - though not ideal when mixing sources.

The makers of the original disc must have messed up somewhere in their workflow and it means there are a lot of frames that look like this:

[img]http://i.imgur.com/ZUak59n.jpg[/img]

Could this be sorted out with avisynth?

I have a fair amount of experience with deinterlacing across a variety of sources, but can’t tell what you are dealing with for certain without access to a short clip… but that ghosting appears to be blended fields (seen on some 24->PAL->23.976 conversions) in which case it is possible to bring back the original cadence (the old RePAL filter for Avisynth + a smartbob filter does the trick) but the footage will still have blurriness. Gets you back to 23.976 without dropped frames though.

In order to fill a 720P 4:3 frame vertically, the original 1080P source only needs to be scaled down to 96%.

I thought I would add that it makes more sense to trim the few lines off the 1080 source to make 720 rather than scaling it down 96%. You gain massively on image sharpness (very slight scaling does the most damage) and only lose 4% on the edges - hardly anything at all given what is already lost to the P&S process!