Decimating exactly to 25 from 29.97 (PAL-sourced - by duplicating frames)

[Alex_ander]

I'm trying to restore a bad PAL=>NTSC conversion (performed by TMPGEnc2.5 as GSpot says). D2v project in AviSynth shows each 6th frame to be a duplicate (with its both fields).
Everything looks like the process in TMPG for PAL=>NTSC conversion was:
separating fields => resizing => interlacing => insertion of duplicates.
If I remember well, TMPG keeps video duration, then the cycle is not 100% regular.
Simple Decimate(6) from Decomb removes the dupes, but it should also remove each 1000th unique frame (as it produces 24.975) - I don't want this.
Also tried TDecimate(mode=2,cycle=6, rate=25), manually came across remaining dupes (unlike with Decomb's Decimate). Didn't try two pass mode though, not sure how to use it.

Is there any smart procedure for doing what I want?
And a related question: is it true that TMPG just separates fields before resizing (doesn't bob, looks like going back to PAL size would be better done for fields then)?

[manono]

Hi-

If you Decimate(6), you get 24.975fps with no dupes. If you use TDecimate(mode=2,rate=25), you get 25fps with dupes every 1000 frames on average. There's no way around that. If it has to be 25fps with no dupes, then:

Decimate(6)
AssumeFPS(25)

and stretch the audio. Or, if for PAL DVD, Decimate(6) (or the TDecimate equivalent) with DGPulldown at the end for 24.975->25fps. I'm assuming you don't want to do anything stupid like blending frames.

I couldn't tell from your post, but is this thing interlaced? Then you have more options. Once you get it to 24.975fps, separate the fields or smart bob it, and then ChangeFPS(50) and reinterlace it. Smart Bobbing is better because you'll be resizing this for PAL, right? And you can do the resize while it's progressive. You'll get duplicate fields every once in a while (every 20 seconds or so) and it should still play pretty smoothly.

I don't (and wouldn't) use TMPGEnc, so I can't help there.

Simple Decimate(6) from Decomb removes the dupes, but it should also remove each 1000th unique frame (as it produces 24.975) - I don't want this.

If the source is NTSC 29.97fps and if every 6th frame is a dupe, with no pattern breaks, then no unique frames will be removed.

[Alex_ander]

Thanks for the answer, manono.

Simple Decimate(6) from Decomb removes the dupes, but it should also remove each 1000th unique frame (as it produces 24.975) - I don't want this.

I couldn't tell from your post, but is this thing interlaced? Then you have more options. Once you get it to 24.975fps, separate the fields or smart bob it, and then ChangeFPS(50) and reinterlace it. Smart Bobbing is better because you'll be resizing this for PAL, right? And you can do the resize while it's progressive. You'll get duplicate fields every once in a while (every 20 seconds or so) and it should still play pretty smoothly.
..........
If the source is NTSC 29.97fps and if every 6th frame is a dupe, with no pattern breaks, then no unique frames will be removed.

It is interlaced and the thing coming from TMPG (that was used by somebody) is that those dupes are simply copies of interlaced frames. After Decimate(6) the fields are separated without visible interlacing artifacts, that is why I assumed the sequence of operations TMPG had done. If I have clean separated fields that had been resized from PAL by TMPG just before interlacing, wouldn't it be better to reverse the resizing process at this point rather than use smart bob?

As far as I can test it manually, the cycle of adding dupes is regular but I can't find out whether there's a break in that addition each 1000 frame. I think it should take place if TMPG keeps constant video duration, then at decimation with regular cycle along with all dupes 1/1000 of original frames are dropped. It's not terrible (I often deal with 24.975) but I wanted to find a perfect solution (if it exists, I mean dropping as many dupe frames as added). Do you have any experience with 2-pass in mode=2 of TDecimate? I might be wrong but in description it looks like a tool for a similar task (at first pass it gathers some info).

In my imagination, the frame sequence in that NTSC video is:
...(5+1),(5+1),(6+0),(5+1),(5+1)...,etc.
where the frames after '+' are dupes and the (6+0) is that (once in 1000) break in addition of dupes at encoding. This could explain why Decimate(6) finds and removes all the dupes, but it is also forced to remove 1 frame from that (6+0).

[manono]

Hi-

In my imagination, the frame sequence in that NTSC video is:
...(5+1),(5+1),(6+0),(5+1),(5+1)...,etc.

Yeah, I was probably wrong with my response in my last post. I apologize. I figured you really wanted to get back to 24.975fps. If it started as 25fps and was converted to 29.97fps by adding dupe frames, then there can't be a regular pattern, and every 1000 frames you'll get that (6+0) thing.

No, I've never used TIVTC's 2-pass. I have, however, used Donald Graft's MultiDecimate's 2-pass many times, but it has a max cycle of 250 frames (if I remember correctly), and you need 1000 frames (?). If you want to use it, in MultiDecimate, anyway, you loaded the first pass script into VDubMod and just hit Play. When done there were the files you needed for the 2nd pass. Perhaps the TIVTC 2-pass method works the same way.

TIVTC's Mode 2 has a look ahead of only 100 frames. That's no good to you, since you need for it to see ahead 1000 (?) frames. You can increase it by adding in M2PA=True. Read the TDecimate doc for more about that.

And here's something I would definitely try, if I were you:

TDecimate(Mode=1,CycleR=166,Cycle=1001)

That will give you 25fps. If there are dupes remaining, it will also mean unique frames were removed, and it didn't work perfectly.

[Alex_ander]

Yes, I'm sure it was previously 25 (not 24.975) since it comes from a French TV broadcast and unlikely the person who plainly fed it to TMPG would previously have done such framerate adjustment.
And your latest suggestion with TDecimate seems to work! So far I don't see any remaining duplicates and have 25 fps at this. Thank you very much, manono!