Avisynth the correct removal of pulldown

[Megafox]

Tell me how to configure the built-in Avisynth filter, or another plugin, to perform the correct removal of pulldown. The built-in filter works correctly only for some video fragments, where 2 frames out of 5 - interlaced, in other places there is even more interlacing. How to make this filter determine the number of interlaced frames in each group of 5, or where to find another plug-in that can do it?

[amaipaipai]

Post a small sample of it.

[hello_hello]

TIVTC is probably the best option. It's quite configurable, but as a rule it works very well straight out of the box. TFM does the field matching and de-interlacing of any combed frames. TDecimate removes the duplicates and/or converts to a constant frame rate.

For a purely telecined source:

TFM()
TDecimate()

For a source that's mostly telecined (film) but contains sections of interlaced video, this converts the lot to 23.976 (the video sections are de-interlaced to 29.97fps and frames are blended/dropped to convert to 23.976).

TFM()
TDecimate(hybrid=1)

For a source that's mostly interlaced video but contains telecined sections, this does it the other way around and converts the whole lot to 29.970fps.

TFM()
TDecimate(hybrid=3)


That's the basics of it.

This is the modern version. I've been using it for quite a while without any problems, but just in case,
here's the original version from somewhere around 2008.

TIVTC.dll includes the above functions.

[Megafox]

Post a small sample of it.

DoubleWeave()
Pulldown(0,3)

[Megafox]

TIVTC is probably the best option. It's quite configurable, but as a rule it works very well straight out of the box. TFM does the field matching and de-interlacing of any combed frames. TDecimate removes the duplicates and/or converts to a constant frame rate.

For a purely telecined source:

TFM()
TDecimate()

For a source that's mostly telecined (film) but contains sections of interlaced video, this converts the lot to 23.976 (the video sections are de-interlaced to 29.97fps and frames are blended/dropped to convert to 23.976).

TFM()
TDecimate(hybrid=1)

For a source that's mostly interlaced video but contains telecined sections, this does it the other way around and converts the whole lot to 29.970fps.

TFM()
TDecimate(hybrid=3)


That's the basics of it.

This is the modern version. I've been using it for quite a while without any problems, but just in case,
here's the original version from somewhere around 2008.

TIVTC.dll includes the above functions.

This plugin does not do what I want, it uses a close search, and eventually gets 1% corrupted and interlaced frames. The most correct way is for groups of frames, but I have both 2/5 and 1/5, how do I remove pulldown for such a video?

[sneaker]

Like amaipaipai said you need to post a sample of your source file if you want decent help. (Cut directly from VOB/m2ts using e.g. DGSplit)

[jagabo]

Pulldown() doesn't adapt to changes in the pulldown pattern.

I have both 2/5 and 1/5, how do I remove pulldown for such a video?

Maybe you just need to adjust the cthresh paramater in TFM(). Or maybe your video was edited, slowed, or sped up after pulldown and can't be restored to 23.976 fps without duplicating/dropping frames. One way around that is to encode at double frame rate:

Interleave(TFM(field=1, pp=1, cthresh=9), TFM(field=0, pp=1, cthresh=9))

That will produce 59.94 fps from a 24.976 fps source. You may need to adjust cthresh. And you may need to call TFM() again to deinterlace any remaining combing. For BFF video reverse the field arguments.

[hello_hello]

This plugin does not do what I want, it uses a close search, and eventually gets 1% corrupted and interlaced frames. The most correct way is for groups of frames, but I have both 2/5 and 1/5, how do I remove pulldown for such a video?

If it can be removed in can be done with TIVTC. TDecimate has several modes. It can run two passes and even create a timecodes file for variable frame rate encoding. Read the TDecimate help file.

In your opening post you had 2/5 sections and sections with even more interlacing, now it's 2/5 sections and 1/5 sections, and we don't even know what the frame rate is. To avoid repeating the question over and over while the rest of gaze into our crystal balls, you'll need to upload a sample.

[johnmeyer]

As others have said, TIVTC is an incredible plugin and can do just about anything, but for difficult video that has different pulldown cadences, you may have to create a script that has more than the usual simple:

TFM()
TDecimate()

that others have already posted.

If you really want to solve this problem, you need to do two things: post a clip (as others have already suggested) and then also post the exact script you are using that, at the moment, doesn't quite work.

I do media restoration for a living and have to use TIVTC almost every week. I've explored just about every setting in both TFM and TDecimate, and have even created external software to take the TFM hints, massage them, and then feed the decimation pattern back into TDecimate.

If the pulldown cadence changes (with mixed film and video being the most obvious and common example), you may need to break the video into pieces and treat each one differently. It would be extraordinarily unusual for the pattern to change within a scene: usually any changes happen at scene changes, when the source for the video changes.

So, post a clip and a script, and when you post the clip, make sure it is long enough to include some of the weirdness, and also make sure it is an original clip, and not something you have re-encoded.

P.S. One "trick" that sometimes works for patterns that seem to change a little is to increase the number of frames considered for each removal. You want to read up on Cycle and CycleR. If you specify a much larger value for Cycle than the usual 5, TDecimate will look at that much larger number of frames before deciding which fields to decimate. The larger Cycle setting will begin to slow down the script, but it can really help with video that has unusual patterns around a scene change, or where the pattern isn't hard-coded, but more variable, something I have often seen with streaming video. When you increase Cycle, you must also increase CycleR so that the "M in N" decimation ratio stays the same.

[manono]

P.S. One "trick" that sometimes works for patterns that seem to change a little is to increase the number of frames considered for each removal. You want to read up on Cycle and CycleR. If you specify a much larger value for Cycle than the usual 5, TDecimate will look at that much larger number of frames before deciding which fields to decimate. The larger Cycle setting will begin to slow down the script, but it can really help with video that has unusual patterns around a scene change, or where the pattern isn't hard-coded, but more variable, something I have often seen with streaming video. When you increase Cycle, you must also increase CycleR so that the "M in N" decimation ratio stays the same.

Yes, many's the time I've used Cycle=10,CycleR=2 or even Cycle=15,CycleR=3. One has to be able to adapt.

But without a sample one can only guess.