twittering and srestore

[martinfrombern]

Hi,
I'm trying to repal an ntsc-ed bbc source. I'm using this script
AssumeTFF()
Yadif(mode=1)
srestore(frate=25)
Everything works well, except there is an annoying twittering/bouncing (as seen when using most average deinterlacers). Fortunately, it's present in only minority of scenes. However, I would still like to remove/reduce it. Any effective way to do it?
I tried other bobbers (e.g. such as yadifmod with nnedi2) but no succeess :( unfortunately using tempgauss or mcbob is imposssible since it's too slow for me (especially considering it's a 1080p content).

[m3mbran3]

You could try TDeint, or TDeint with NNEDI2.

[shoopdabloop]

TDeint+NNEDI2 would be a good choice.

or you could try a slight vertical blur.

[Didée]

Most probably all those will show the same behaviour. Instead, give a try on e.g.

AssumeTFF()
# Yadif(mode=1) # temporarily deactivated
bob(0,0.5) # use this to see more clearly if the trick below helps
ReduceFlicker(strength=1)
srestore(frate=25)

- which is a recommendation I gave long ago, and the reasoning behind has not changed. IIRC, it even can be found in the readme of Srestore.

BTW, the problem is spread between
1) bob filters that do leave bob shimmer behind
2) blend conversions have a repeating phase shift (well, a sort of)
3) on fieldblended-converted sources, the ReduceFlicker-trick is more or less failsafe (but only on fieldblended stuff)


And lastly: You should not use TempGaussMC or MCBob on fieldblended sources, indeed. However not because of the speed, but for technical reasons. Motion compensated bob filters will most likely *fail* on fieldblended stuff, because the ME can not find much useful when searching between a blended and a not-blended field. So at best you'd get the pure spatial interpolation, including the typical twitter; at worst you'd get (the risk of) artifacts due to failed mocomp. In practive, you'd get a mix of both.

This, too, has been mentioned several times in the past.

[martinfrombern]

@m3mbran3 and shoopdabloop: tried that before. no success :(

@Didée: thanks for interesting info. unfortunately, your trick doesn't help. i get the same twittering effect as before :( any other way to eliminate it? or there is no help and i just have to leave it as it is atm?
EDIT: your answer also raises one more question... i've noticed that only tempgaussmc and mcbob work relatively good with vertically scrolling credits (at the end) in my source (they are not perfect, but are considerably better than others). so what to use insted of them for this part?

[thewebchat]

Why not just use your motion-compensated deinterlacer for the credits and your standard dumb deinterlacer elsewhere.

[martinfrombern]

Why not just use your motion-compensated deinterlacer for the credits and your standard dumb deinterlacer elsewhere.because of reasons mentioned before
1) it's too slow, even for a small amount of source (such as ending credits). It would increase encoding job for at least one day
2) (and more important) as didee said, there is a risk of other artifacts, especially if you take into account that there is action/movement behind those scrolling credits (instead of usual black background).