logo Sign In

Inverse Telecining: Best method for restoring original frame rates

Author
Date
Author
Time
Post link
Well, the title says it all. Sorry this is a redundant question, but the search on these forums isn't good enough to turn up where this has been asked before.

What is the best/cleanest way to restore the original 24 frames by inverse telecine?
I've heard good things about VirtualDub(Mod's) internal alorithm, but I had so-so luck with it when I tried it before.
Every guide online suggest different methods and everyone has a different opinion what's best.

I know a perfect method is impossible because some original frames only exist as fields split between different frames.

I've seen some wonderful work done on Star Wars restorations, so I know there are people doing this right.

Edit: I saw that ADM recommends:
Telecide(order=1,guide=1).Decimate()
How effective is this?

Dr. M

Author
Time
Post link
Very effective.

in avisynth 2.5

Telecide(guide=1,post=2,vthresh=25)
Decimate()

is what I use as a base to IVTC from 29.97 to 23.97/24/25

adding

AssumeFPS(24)

Would speed it to 24fps..

Author
Time
Post link
Is that any different from the way ADM's suggestion works?

Also I mis-spoke. I meant 23.976 since I'm keeping an NTSC format.
Will assume(23.976) be necessary?

Dr. M

Author
Time
Post link
(I moved this thread to technical discussions.)

What's your source?

A DVD source with well-behaved pulldown is probably best handled using the "force film" option in DGIndex.

Any other source, if the pattern is constant, can be IVTCed very quickly by
doubleweave()
pulldown(a,b) # a and b need to be chosen manually.

Strange/inconsistent pulldown patterns may need to be handled by an automatic AVISynth filter such as Decomb or TIVTC, but this method is obviously slower because of the field matching algorithms in use.

You do not need assumefps(23.976), because the 29.97 fps video is decimated by 4/5 during IVTC to get 23.976.

Guidelines for post content and general behaviour: read announcement here

Max. allowable image sizes in signatures: reminder here

Author
Time
Post link
No my way is basically the same, I think ADM's is for an older version of avisynth is all - I dont think 'order=' is used anymore on newer versions of decomb.

Moth3rs Doubleweave method is also great because mathmatically it can't go wrong, very good for HD material (for instance) with an obvious 2:3 pulldown/5 frame repeat..
Author
Time
Post link
Thanks for further proving I have no idea what I'm doing. IVTC makes me feel stupid (that's why I usually avoid it.)

a and b need to be chosen manually

Um what?
Am I correct that with doubleweave, almost every frame has to be examined and if the pattern shifted you have to add a new line for that section of film?


The source is a laserdisc capture (2 sided).

I also have a second question (yup, even though the first is still hazy to me):
I tried to contact Citizen regarding his upsizing method to hi-def Xvid. Since I'm looking to turn a letterbox source to anamorphic I'm obviously scaling upward. Normally I'd just plug in a lanczos4 filter, but he did such a nice job I was hoping someone knew what he did exactly beyond his vague "multiple resizes with different filters".

Dr. M

Author
Time
Post link
Originally posted by: Doctor M
Thanks for further proving I have no idea what I'm doing. IVTC makes me feel stupid (that's why I usually avoid it.)

a and b need to be chosen manually
Um what? There are five different combinations for a and b. Two of the five will work equally well for normal 3:2 pulldown. To find which values to use, use the ShowFiveVersions script.
Originally posted by: Doctor M
Am I correct that with doubleweave, almost every frame has to be examined and if the pattern shifted you have to add a new line for that section of film?


The source is a laserdisc capture (2 sided).
Unfortunately, for laserdisc captures it might not be a case of choosing values that work through the whole capture. For example, Pulldown(2, 4) might work at the beginning of the file, but at some scene change the pattern may change and Pulldown(0 ,3) is required for a different portion. This is where manual methods start to become tedious, and automatic filters start to look more attractive.

Originally posted by: Doctor M
I also have a second question (yup, even though the first is still hazy to me):
I tried to contact Citizen regarding his upsizing method to hi-def Xvid. Since I'm looking to turn a letterbox source to anamorphic I'm obviously scaling upward. Normally I'd just plug in a lanczos4 filter, but he did such a nice job I was hoping someone knew what he did exactly beyond his vague "multiple resizes with different filters".
Lanczos should be fine for a letterboxed - anamorphic upsizing. I believe Citizen used an implementaion of Edge Directed Interpolation for upscaling to HD resolution. There's a couple of filters mentioned in this Doom9 thread:
http://forum.doom9.org/showthread.php?t=72174

Guidelines for post content and general behaviour: read announcement here

Max. allowable image sizes in signatures: reminder here

Author
Time
Post link
Thanks for the input.
I'm waiting on the captures, but I'll see what works and let you all know.

Edit: Reading the thread on EDI resizing, it looks like most of the discussion is experimental/theoretical.

Edit2: Has Citizen left the forum? I've seen nothing from him in awhile, and he doesn't seem to answer emails to the address he has in his sig.

Dr. M

Author
Time
Post link
Citizen doesn't post here much any more, but he is still around.

Have you tried the address at the top of this page?

Guidelines for post content and general behaviour: read announcement here

Max. allowable image sizes in signatures: reminder here

Author
Time
Post link
I quick test of one chunk of about 10 minutes looks like I can get away with:
doubleweave()
Pulldown(1,3)

Is it common to see light/mild ghosting on occasion with those filters? I'm not talking interlace artifacts, it almost looks like a temporal artifact of some sort.

Dr. M

Author
Time
Post link
No, because this method is a simple re-construction of the original frames. It is completely non-destructive, because there is no post-processing involved.

Any temporal artefacts you are seeing will be present in the original frames.

Guidelines for post content and general behaviour: read announcement here

Max. allowable image sizes in signatures: reminder here

Author
Time
Post link
Know anything about chroma artifacts? I start a new thread with a question about that.

Edit: Oops, didn't merge the full project. Still the telecine seems consistent across the whole film (I trimmed a few black frames between the side changes so the telecines match up for both sides).

Still, thanks for the input guys, completely bang on advice.

Dr. M

Author
Time
Post link
New question asked... and withdrawn.

Dr. M