this retail MOD dvd has significant judder, apparently made from a PAL source for Ntsc. It still has the PAL sped-up runtime and audio pitch. I wanted to re-encode for dvd video using HCenc, correcting for judder and PAL. A little disorienting is the huge black bars because it was encoded to 16:9 (which I'll crop later).
Stepping thru in AvsP with separatefields, it looks like 2:5 pulldown, but simply IVTCing produces jerky video.
Yadif/Srestore also makes jerky video.
TFM().TDecimate(Mode=2,Rate=25) seemed great - I assume it should create a field with a duplicate, but I found several single fields. When encoding, those areas of single fields looked like combing. I'm uploading a sample from the dvd. When the guy is talking at about 17 seconds in, that's where (with my script) I see several separate fields without dupes, which I guess is what's creating the combing when I re-encode (whether 25 fps or 23.976 with pulldown). Any help much appreciated.
+ Reply to Thread
Results 1 to 20 of 20
Yep. TFM(cthresh=2).TDecimate() alone might do it, as shown .
The movie is progressive, with 1 dupe frame every 6 frames. Maybe a longer clip will reveal periodic changes in that pattern.
[EDIT] changed to show jagabo's idea more clearly.
Last edited by LMotlow; 20th Nov 2015 at 11:39.- My sister Ann's brother
I meant he should add cthresh=2 to TFM. He still need to decimate. Either with what he has or with TDecimate(Cycle=6, CycleR=1).
OK,got that. They both worked over here. First time I've seen that kind of conversion. I usually see something more complicated.
I got the attached mpg with TFM().TDecimate(Cycle=6, CycleR=1), 3:2 soft pulldown added. Jagabo's version gave me the same thing.- My sister Ann's brother
beautiful, jagabo. Thanks for saving my hash yet again. I stayed with the TDecimate(Mode=2,Rate=25), seemed perfect for that sample, so I didnt try the cycler script - is there a preference of one over the other?
TDecimate(Mode=2,Rate=25) will leave you with 25 fps. TDecimate(Cycle=6, CycleR=1) will leave 24.975 fps. Which is correct will depend on how the video was frame rate converted from PAL to NTSC. The basic technique used was to duplicate every 5th frame.
1) If that was done directly from the 25 fps PAL source it would give 30 fps. You'd have to then discard every every 1001th frame (or skip the corresponding duplication) to get back to to 29.97 fps. The equivalent commands in AviSynth would be:
# starting with a 25p source ChangeFPS(30000,1001)
2) Or the PAL source may have first been slowed to 24.975 fps, then every 5th frame duplicated to make 29.97 fps. The equivalent AviSynth commands:
# starting with a 25p source AssumeFPS(25000/1001) SelectEvery(5, 0,1,2,3,4,4) # or ChangeFPS(30000, 1001)
The advantage of #1 is that the run time isn't changed so audio doesn't have to change. My guess is that's what you have.
The dvd was a double-feature, and unfortunately the other title seems to be PAL-modified-for-Ntsc playback; I'm going by the runtime being about 4% shorter than what's expected, and the audio sounds higher pitched than my vhs-sourced avi. There's much less judder, possibly because they used blends instead of dupes, and I havent been able to unblend it correctly.
Last edited by spiritgumm; 20th Nov 2015 at 19:55.
A PAL video was converted to NTSC with 3:2:3:2:2 pulldown. Then the video was blend deinterlaced for some reason. That leaves 2 frames out of every 6 blended. Unfortunately, one of the original film frames can only be found in those two blended frames. It's hard to reconstruct it but this works pretty well:
FixBlendIVTC() TDecimate(Cycle=6, CycleR=1) # or mode=2, rate=25
thanks jagabo, I'll check it out. Btw, these movies are on the "Dark Intruder/Night Walker" MOD dvd from TCM. Universal keeps taking PAL versions of their films and modifying them for the Ntsc market. Really lame.
I re-encoded the first movie, and seems like every 30 seconds or so, I get duplicate frames (and possibly dropped frames). If adjusting cthresh will help, should I go even lower? If the range is -1 to 255, it's almost maxed out (right?).
I don't think a lower cthresh will help. That's the threshold value used to determine if the frame is already progressive. All it changes is how many frames with a tiny bit of combing get through TFM.
I suspect the problem is in TDecimate. Especially with mode 2 it can be hard to determine exactly which frames should be removed. I'd try the cycle/cycler method and see if that works any better.
the cycler method works better (for both videos). Still some issues. I noticed in at least one spot that the duplicate frame occurred on the 7th frame, so unfortunately the 6th frame is removed and the video appears to jumps forward. I dont know if that's the general problem.
In the other video with blends, some decimation issues in a different sample. Stepping thru using just fixblendivtc, most dupes were of the 5th frame, but some dupes were of the 4th or 6th frames.
If there are dupes left over I wouldn't worry about it. If there are jumps from missing frames then you have a problem. Duplicate frames are better to have than missing frame.
In the first video, in the clip I was using, there was no shorter cycle. It was 5 frames with 6th frame dupe, until the single cycle with 6 frames and dupe. Using TDecimate(Cycle=18, CycleR=3) the missing frame happens slightly earlier than with cycle=6, so not an improvement.
Cycle=18 seems to improve the blended video, although I'm not sure how in the clip with the shorter cycles of 4th and 6th dupes. I havent encoded for 23.976 and watched on TV, so I dont know how bad the light-edging looks.
What about using Srestore with double-blend removal and separate decimation? I've read over the notes, but I dont think I'm setting it up correctly.
You may wind up with more dupe frames, but the main goal is to not lose any frames. Override files can be used as well, but I've always found them to be more trouble than they were worth.
...although I'm not sure how in the clip with the shorter cycles of 4th and 6th dupes.
Last edited by manono; 21st Nov 2015 at 18:20.
At this point, I think your Cycle=18 might actually be better for the first movie, at least in the scenes I've watched. A higher cycle seems to eliminate the missing frames, but I notice the dupes more. One of those cost/expense ratios.
The 2nd movie (the blended one) has alot of aliasing/combing. It's not noticeable in the short dark clip I uploaded, so here's another sample. Regardless of which cycle (6 or 18) to use, I'm wondering if the combing can be minimized as it was with TFM in the first movie.
What combing? Unless I'm missing something, there's no interlacing at all and it's progressive. It looks to me like it was hard telecined from PAL's 25fps to NTSC's 29.97fps and then blend deinterlaced. The result is two blended frames in every six-frame cycle. Am I wrong about that?
ExBlend claims to be able to unblend such PAL2NTSC sources in a way that SRestore can't.
The filter will also deal with similarly blended 25 FPS to 30 FPS video in which case
the blending would be as below:-
A B C D E # 25 FPS Source (5 blocks of 5 frames = 25 FPS)
A B cb cd D E # Blended to 30 FPS (5 blocks of 6 frames = 30 FPS)
Again, it may be slowed to 29.97 FPS. In this case there will be a repeated sequence
of 4 good, 2 blended frames.
I see jaggies in the clip on my monitor, and on my blu-ray player. When I process it with jagabo's script below, it's just as apparent.
Yes, there are aliasing artifacts from a poor deinterlace or resize. Try using Santiag() or some other antialiasing filter. Also consider dehalo_alpha() for the oversharpening halos.
Coincidentally, Exblend worked for that "Car" clip. It didnt work for a longer clip, nor for the earlier posted clip with the clock.
So it's back to jagabo's script, with antialiasing.