Avisynth rpow2 upscale, or not?
 

@lordsmurf--
In another post from last Fall, you suggested I give 'Avisynth nnedi3 rpow2' a try for interpolating the captured Hi8 footage rather than using anything by Topaz Labs (Gigapixel AI or VideoEnhance AI).

Well, 50 some odd tapes later, I'm finally in a position to investigate the best method for upscaling some of this SD footage to HD.

Googling the avisynth software yields precious few results. 107 hits to be exact. I can't tell what platform / environment this software runs on / under. Could you illuminate?

Moreover, the Wiki for nnedi3 refers to it as I realize you were directing me at nnedi3_rpow2 (which enlarges images by a power of 2), but why would anyone want to use nnedi3 for deinterlacing captured footage? Moreover, why would one want to throw one field away and construct a progressive scan from only one field??

I'm using Adobe Premiere Pro to nicely deinterlace the raw footage by exporting as Progressive (by changing the Field Order parameter from Lower to Progressive).

Trying to figure out if / how I would use nnedi3_rpow2. Guess I should remind I'm on MacOS.

If you use QTGMC (in AVISynth) to de-interlace your Hi8 captures, you may well see better results than you will get out of Premiere.

Granted it is a lot more involving to set up, but there is at least one very helpful video on YouTube about Deinterlacing with AVIsynth and QTGMC.
That will help you get set up.

Thanks. You might be referring to this youtube video by Swan: https://www.youtube.com/watch?v=C4PyyQoz6eo. I'll finish watching this by week's end.

I guess it's pretty clear that Avisynth is a Windows platform capability.

You can use vapoursynth instead on macos. Probably the easiest way to use qtgmc tho is using hybrid, which can do most of the work for you.

Yes that's the one.
Although first I followed the one he did 2 years earlier Installing 32-bit and 64-bit AVISynth+, QTGMC and FFMPEG side-by-side.
Then followed that late 2020 video to add VirtualDub2 etc. Slow videos but very thorough. If you can install on a Windows system it's best to ensure your Windows User Account (UAC) has Administrator privileges, that should help the install go smoothly.

On MacOS I have no idea if that can be set up. Best as hodgey says try hybrid that is well respected on this forum too.

I use Nnedi3 to apply progressive filters on interlaced videos: create a progressive video for filtering, then revert to interlaced video as output. It allows (ideally) a "lossless" operation.

<open interlaced video>
<Lossless Bobbing deinterlacer = Nnedi3>
<filtering = TemporalDegrain2: Denoiser, Temporal filter with motion compensation>
<filtering = LSFmod: Sharpener>
SeparateFields()
SelectEvery(4,0,3)
Weave()

Hmm. I don't perfectly understand your answer but recognize that I am only at the start of this learning curve. It still doesn't make sense to me why one would throw away one field in an interlaced video in order to transform (deinterlace) to progressive.

nnedi3 isn't throwing out fields here. The rpow2 isn't the same.

And QTGMC is a complex script that just leverages nnedi's (or others). I suggest QTGMC, not a nnedi alone.

Fields are 50% of frame, at intervals, temporally out of sync in whole. It's not complete data. There are different ways to approach deinterlacing, an nnedi is at all the worst of them.

The statement "throws away one field" is confusing, generating the idea that something is "lost".

from Dideè, author of MCBob and TGMC, becaming later QTGMC:

A same-framerate deinterlacer will ask you for a "field=top/or/bottom" parameter, defining which of both fields will be thrown away & interpolated: Same framerate = one field doomed.
A bobber does the same, but after each frame inserts another frame that is based on the previously thrown-away field: Framerate doubled by not dooming one field.

Nnedi3 is just an interpolator and is not a smart bob on its own, It would never be able to beat a motion compensated or motion adaptive bobber. If you want to deinterlace use QTGMC.

I use Needi3 with the option field=-2 (double frame rate) to act like a bobber (but with better quality than standard Bob()) in order to build a "progressive" video from interlaced to make the filtering operation more effective. Then I go back to interlaced, removing interpolated fields.

interlaced frames video
frame1 frame2 frame3 frame4 frame5 frame6 frame7 frame8
A..............C
b..............d

Bob() or Nnedi3(field=-2)
frame1 frame2 frame3 frame4 frame5 frame6 frame7 frame8
A...........B'.........C .........D'
a'..........b..........c'..........d
x' and X' represents scanlines interpolated from x and X

SeparateFields() assuming TFF
frame1 frame2 frame3 frame4 frame5 frame6 frame7 frame8
A..........a'..........B'.........b.........C..... ....c'...........D'.........d

SelectEvery(4, 0, 3)
frame1 frame2 frame3 frame4 frame5 frame6 frame7 frame8
A..........b...........C........d

Weave()
frame1 frame2 frame3 frame4 frame5 frame6 frame7 frame8
A...........C
b...........d

P.S. I could also use QTCMG to deinterlace, but QTGMC is not really lossless and it performs denoising on its own. I obtain better results with TemporalDegrain2.
(BTW some people is now working on integrating TD2 in QTGMC because they share same motion compensation approach ;))