Modified Restore24 results + word about DGPulldown

[mantis2k]

In another thread, the creator of Restore24 was warning me about using his deinterlacer (poor Field Blended to smooth Progressive) for outputting frame-rates other than 23.976--it's designed for bad 35mm film transfers.

Despite what Didee said, which made sense, I am unable to spot any noticeable flaws! Below are results from various tests I carried out on my source--probably the worst Field Blended source ever posted on Doom9--not even SRestore could handle it.

Instead of carrying out industry standard PAL speed-up, I cheated with Restore24, maintaining original running time. Perhaps you don't think 23.976 > 25.000 should be very noticeable by using trickery? What if told you I can't spot any flaws using Restore24 to deinterlace Field Blended to Progressive--AND--go from 23.976 > 29.970!? :)

Restore24(r24fps=23.976,numr=2997,deno=6250,useLL=true,stringuent=false,fulltriples=false,r24deint=1,fra=1,frr=1,edgetype=0,edgemode=2,edgeeval=2,r24size=.75,r24ysize=0.75,nr=-5,ldmp=32.0,debug=0)
Download... (http://www.rarekungfumovies.com/speedup2.mpg) (Restore24 + PAL speed-up)

Restore24(r24fps=25.000,numr=3125,deno=6250,useLL=true,stringuent=false,fulltriples=false,r24deint=1,fra=1,frr=1,edgetype=0,edgemode=2,edgeeval=2,r24size=.75,r24ysize=0.75,nr=-5,ldmp=32.0,debug=0)
Download... (http://www.rarekungfumovies.com/trickery2.mpg) (Restore24: 23.976 > 25.000)

Restore24(r24fps=29.970,numr=3746,deno=6250,useLL=true,stringuent=false,fulltriples=false,r24deint=1,fra=1,frr=1,edgetype=0,edgemode=2,edgeeval=2,r24size=.75,r24ysize=0.75,nr=-5,ldmp=32.0,debug=0)
Download... (http://www.rarekungfumovies.com/NTSCTrickery2.mpg) (Restore24: 23.976 > 29.970)

A word about DGPulldown: using this application to carry out a 3:2 pulldown from 23.976 > 29.97 may seem good when played back on PC, but troubles start the moment you do something else:
a) processing video or cutting a clip causes playback of newly saved video to become jerky.
b) playback on standalone DVD player plays back jerky like (a).
Note: DVDMaestro imports DGPulldown material as "dropped frames".
To be honest, I would not recommend pulling down the frame rate from my experience of PAL to NTSC conversions..
http://neuron2.net/dgpulldown/dgpulldown.html

How would you go from 23.976 to 29.97? Perhaps other software does better 3:2 pulldown, or maybe Canopus Procoder might be decent for this? All I know is: I've got no problems using Restore24 for deinterlacing, frame-rate conversion and keeping original running time.

[Didée]

Harrrrr. How to make it compact ...

The true underlying original framerate is 23.976. Period.
Let me show you with a "clean" case of fieldblended PAL what happens when you do what you have done:

- When restoring fieldblended PAL to 23.976 directly, then you get THIS (http://www.mediafire.com/file/tjmz2mwmznd/forest_PAL-blend-restore_to-23976.avi).

- When restoring fieldblended PAL to 25.000 directly, then you get THIS (http://www.mediafire.com/file/uwrhmmyvumf/forest_PAL-blend-restore_to-25000.avi).

You see the difference, I must not explain that. (If you want to try by yourself, here's the source (http://www.mediafire.com/file/m0olwtvw2iz/_wald_short.demuxed.m2v).)

With your source, it's basically the very same. The difference is, your source is so much broken that no deblender can make it *really* smooth playing. Also R24 makes a good amount of mis-guessings, there is quite some blending going on, and the result is smooth enough to be acceptable and watchable, but in no way it is really fluid.

Believe me, I *can* see the motion breaks in your "trickery" 25->25 conversion. I agree that they're not very obvious, since they get mostly lost in the overall not-quite-smoothness of the result. It's not easy to see, but it is there.

Now, point is you did something that's technically very wrong. The lucky point is that the error is hardly visible. You can't hear your mate talking when an airplane is starting a few meters away, that's not surprising.

But really, just because the error is hardly visible because of "lucky" circumstances, you shouldn't brag about what a cool method you found. You did wrong, and by chance one can hardly see the disaster. That's all.

[mantis2k]

Thanks for taking to time to explain and provide examples. I know what to look for now. However, I still cannot spot any flaws. Even if it's not really smooth, at least it doesn't have those periodic pauses as shown in your example.

Do you think your other filter--TempGaussMC--would be able to deinterlace my field blended source any smoother? I'm still very happy with Restore24, yet I'm quite a fussy person.

[mantis2k]

BTW, from my latest test, DGPulldown is actually good for 3:2 Pulldown, otherwise going from 23.976 to 25 does come out jerky.

[scharfis_brain]

3:2 pulldown judder vs. 24p to 25i pulldown judder is a case of personal preference.

I prefer two stutters per second (24p to 25i) over twelve stutters per second (3:2 pulldown)