Fun with Bill Nye end credits and deinterlacing

[Katie Boundary]

NNEDI3:

https://imgur.com/Q0Ryyp7

YADIF, mode 1:

https://imgur.com/C2j6ey9

YADIF, mode 3:

https://imgur.com/a9LQM3q

Using the edeint parameter to pull deinterlaced pixels from NNEDI3 didn't help (in fact, that was how I learned about the problem). The issue is with the temporal awareness. I ended up using NNEDI3+YADIFmod2 for most of the episode and then switching to NNEDI3 alone for the end credits.

[kedautinh12]

Can't see nnedi3 with undefined image

[johnmeyer]

Not much use without being able to see the original. :)

The reason for wanting to see the original video is that end credits which stroll down the screen often have issues that must be dealt with before deinterlacing. In particular, if the scroll on interlaced video is not done correctly by the producer of the video, you end up with the tops and bottoms of letters not being at exactly the same position relative to the scan line or pixel row as it was in the previous frame. Put another way, to get a really smooth scroll, the speed of the scroll needs to be mathematically tied to the number of vertical scan lines and the frame rate. If it isn't, you will get some nasty twitter in the original interlaced video when viewed on an old-fashioned interlaced CRT monitor, and the problem gets magnified if you then apply deinterlacing.

So, like I said, you need to post the original, and you need to post actual video rather than stills in order to see if your efforts are providing a real improvement.

[Katie Boundary]

Bloody hell... stupid goddamn imgur...

Well, I didn't have my original script or d2v file anymore, but I did have the final AVI file, so I've posted the equivalent frame from it. The point is that it looks fine, or at least the text does.

[flossy_cake]

end credits which stroll down the screen often have issues that must be dealt with before deinterlacing.

QTGMC's InputType=2 (progressive repair mode, not deinterlacing) can smooth it out after the fact:

https://b.l3n.co/i/sy0QZv.png

For InputType=2 I believe QTGMC uses NNEDI3 interpolations to do the smoothing between fields.

Original clip: https://drive.google.com/file/d/1lKtN3ZQ36FwEgqr98cDpLV5C7vyl_Wv9/view?usp=sharing


YADIF

I personally don't like YADIF's field interpolation. I would even go as far as saying it is "wrong". Even BWDIF's field interpolation is not real field interpolation I don't think... it seems to do a blend with the weaved version of the clip or something, I'm not really sure. TDeint interpolation is a bit cleaner:

https://c.l3n.co/i/sy0VsC.png

...but TDeint is a bit too CPU heavy in conjunction with QTGMC so I prefer BWDIF+QTGMC(InputType=1/2).