Types of Pulldown

[Megafox]

I decided to start creating promised AviSynth plugin for manual IVTC by allocating clip boundaries in which cadence changes, removing Pulldown cadence in this clip, replacing lost fields with a copy of previous frame, duplicating frames (+1 to frame count), removing duplicates (manual decimating, -1 to frame count), checking for out of sync, etc. A little later after creating plugin, I will write a manual: how to determine interlace frames and cadence, which is used throughout video in cropped form (duplicate detection in places with small movement, detection by noise, detection by resolution drop in one frame, detection with invert, etc.). To implement support for all types of Pulldown, I need to know how they are arranged, using scheme. DGPulldown is crashing.
In Internet, I found information only about:
1. differences between 3:2, 2:3, 3:2:3:2, 2:3:2:3
2. scheme for 2:3:3:2, but I'm not sure that scheme is not confused with 3:3:2:2 or 2:2:3:3
3. 2:2:2:2:2:2:2:2:2:2:2:3 (12:1, 24:1), described only in words, I still do not understand how it works: 11p frames and 1i duplicate or 12p frames and 1i duplicate
For remaining cadences I did not find description or scheme in Internet.
Tell me any site or pdf, where there is a scheme for following types of Pulldown:
2:2
3:3
2:3:3:2
3:3:2:2
2:2:3:3
2:2:2:4
2:2:3:2:3
2:2:2:2:2:2:2:2:2:2:2:3 (24:1, 12:1)
And explain to me what mean numbers 2, 3, 4, how are composed these titles (2:2, 3:3, etc.)?

[veresov]

You say that DGPulldown is crashing. Can you please link me to your stream and give your settings so I can try to duplicate it and fix it if necessary.

[jagabo]

It's very simple: Each number represents one film frame and the value indicates the number of video fields over which the film frame is displayed.

2:3 pulldown means the first film frame is displayed for the duration of 2 video fields, the next film frame is displayed for the duration of 3 video fields. So 2 film frames becomes 5 video fields (2.5 video frames), then the pattern repeats. The repeating pattern means 2:3 pulldown is exactly the same thing as 2:3:2:3 pulldown. The reason it's sometimes called the latter is to give a whole number of video frames -- ie, 4 film frames becomes 10 video fields (5 video frames). It's especially useful to think of it this way when it comes to inverse telecine -- since you are dealing with an integer number of video frames.

2:3 and 3:2 are essentially the same thing. It's just a matter of where you start. In practice, when you capture video you have no control over the starting frame so a movie broadcast with 2:3 pulldown may appear as 3:2 pulldown in your cap.

Likewise 2:3:3:2, 3:3:2:2, and 2:2:3:3 are all essentially the same thing, differing only in where you start. And again, when you capture you have no control over which frame you start so you may get any of those patterns in the cap.

2:2:2:2:2:2:2:2:2:2:2:3 means the first film frame is displayed for the duration of 2 fields, the next for 2 fields... until frame 12 where a film frame is displayed for the duration of 3 fields. Then the sequence repeats. This converts 24 film frames to 50 video fields without changing the running time (no PAL speedup or pitch change). When this is captured and packaged as frames you get progressive looking frames for half a second followed interlaced looking frames for half a second.

You should be able to figure the rest out on your own now.

[Megafox]

It's very simple: Each number represents one film frame and the value indicates the number of video fields over which the film frame is displayed.

2:3 pulldown means the first film frame is displayed for the duration of 2 video fields, the next film frame is displayed for the duration of 3 video fields. So 2 film frames becomes 5 video fields (2.5 video frames), then the pattern repeats. The repeating pattern means 2:3 pulldown is exactly the same thing as 2:3:2:3 pulldown. The reason it's sometimes called the latter is to give a whole number of video frames -- ie, 4 film frames becomes 10 video fields (5 video frames). It's especially useful to think of it this way when it comes to inverse telecine -- since you are dealing with an integer number of video frames.

2:3 and 3:2 are essentially the same thing. It's just a matter of where you start. In practice, when you capture video you have no control over the starting frame so a movie broadcast with 2:3 pulldown may appear as 3:2 pulldown in your cap.

Likewise 2:3:3:2, 3:3:2:2, and 2:2:3:3 are all essentially the same thing, differing only in where you start. And again, when you capture you have no control over which frame you start so you may get any of those patterns in the cap.

2:2:2:2:2:2:2:2:2:2:2:3 means the first film frame is displayed for the duration of 2 fields, the next for 2 fields... until frame 12 where a film frame is displayed for the duration of 3 fields. Then the sequence repeats. This converts 24 film frames to 50 video fields without changing the running time (no PAL speedup or pitch change). When this is captured and packaged as frames you get progressive looking frames for half a second followed interlaced looking frames for half a second.

You should be able to figure the rest out on your own now.

Did I write it correcrly? Thanks.
2:2 - AA BB
2:3 - AA BB B|C CD DD
3:2 - AA AB B|C CC DD
2:3:2:3 - AA BB BC CD DD
3:2:3:2 - AA AB BC CC DD
3:3 - AA AB BB
2:3:3:2 - AA BB BC CC DD
3:3:2:2 - AA AB BB CC DD
2:2:3:3 - AA BB CC CD DD
2:2:2:4 - AA BB CC DD DD
2:2:3:2:3 - AA BB CC CD DE EE
2:2:2:2:2:2:2:2:2:2:2:3 - AA BB CC DD EE FF GG HH II JJ KK LL L|M MN...

[jagabo]

Did I write it correcrly? Thanks.
2:2 - AA BB
2:3 - AA BB B|C CD DD
3:2 - AA AB B|C CC DD
2:3:2:3 - AA BB BC CD DD
3:2:3:2 - AA AB BC CC DD
3:3 - AA AB BB
2:3:3:2 - AA BB BC CC DD
3:3:2:2 - AA AB BB CC DD
2:2:3:3 - AA BB CC CD DD
2:2:2:4 - AA BB CC DD DD
2:2:3:2:3 - AA BB CC CD DE EE
2:2:2:2:2:2:2:2:2:2:2:3 - AA BB CC DD EE FF GG HH II JJ KK LL L|M MN...

Yes. Maybe the easiest way to think of this is to first duplicate film frames to to the field rate, then select alternating top and bottom fields from the those frames:

So for 2:3 or 3:2 pulldown:

# 24 frame per second source to 60 field per second video
ChangeFPS(60) # duplicate frames
AssumeTFF() # make top field first video
SeparateFields()
SelectEvery(4,0,3) # select alternating top and bottom fields, 60 fields per second
Weave() # 30 frames per second

Remember, when you watch interlaced video on a CRT you never see an entire frame. You see a sequence of fields. Actually, you don't even see entire fields as the phosphors decay very quickly, over the time span of a scan line or so. See this high speed video of a typical CRT TV being drawn:

https://www.youtube.com/watch?v=3BJU2drrtCM

Of course, all this happens too fast to actually "see". All you notice is a slight flicker.

[johnmeyer]

For hard-coded pulldown, the pattern can be anything, although 3:2 cover 99% of the material most people will see. When I used to make DVDs from amateur film, and that film was shot at 12, 15, 16, or 18 fps, I created the appropriate field-by-field pulldown using AVISynth and, more recently, simply using Vegas with "smart resampling" disabled. When I've later had to IVTC some of this material, I use this AVISynth script:

tfm(display=false) 
tdecimate(display=false,mode=0,cycleR=2,cycle=5)  #18 fps  # 3:3:4 field count
#tdecimate(display=false,mode=0,cycleR=7,cycle=15)#16 fps

[ConsumerDV]

I noticed excessive judder on a recent TV show grab, checked the pulldown, and it turned out to be 2:2:2:4 - why would they do that instead of 2:3? Ugh.

Also, I just want to mention that pulldown patterns are used for progressive-scan as well. The only difference is that complete frames are used instead of fields. So, when they say that yay progressive scan will fix interlacing artefacts... yes, but it will not fix pulldown artefacts. A perfect TV set needs to have a buffer long enough to store the whole pulldown sequence, then it should re-arrange frame repeating, in my case converting 2:2:2:4 into 2:3. Even better if it converted it into 3:3 and switched the panel into 72 Hz mode.

Somewhat related: does anyone know, what software did hometheaterhifi use for listing flags, like here for the Beautiful Mind film?

[Brad]

I noticed excessive judder on a recent TV show grab, checked the pulldown, and it turned out to be 2:2:2:4 - why would they do that instead of 2:3? Ugh.

A pulldown sequence with no odd numbers means there's no interlacing. "2:2:2:4" isn't really a thing; it's just what you get when you deinterlace 2:3 to 29.97fps instead of handling it properly.

2:3 - AA BB B|C CD DD

Let's take just the top fields: A B B C D
Let's take just the bot fields: A B C D D

Do you see how in either case, you get a 29.97 output with jerky 1-in-5 duplication instead of spreading the judder more evenly across 59.94?


---


EDIT: I've never seen this thread before, but the OP in 2018 had a similar idea that I had in 2007: IVTC based on scene changes - has it been tried? (Doom9 forum thread)

I decided to start creating promised AviSynth plugin for manual IVTC by allocating clip boundaries in which cadence changes, removing Pulldown cadence in this clip, replacing lost fields with a copy of previous frame, duplicating frames (+1 to frame count), removing duplicates (manual decimating, -1 to frame count), checking for out of sync, etc. A little later after creating plugin, I will write a manual: how to determine interlace frames and cadence, which is used throughout video in cropped form (duplicate detection in places with small movement, detection by noise, detection by resolution drop in one frame, detection with invert, etc.). To implement support for all types of Pulldown, I need to know how they are arranged, using scheme.

[ConsumerDV]

A pulldown sequence with no odd numbers means there's no interlacing. "2:2:2:4" isn't really a thing

Why is it not a thing? Here:

Top fields: A B C D D
Bot fields: A B C D D

For 60p just convert fields to frames: A A B B C C D D D D This is the reason for excessive judder.

It could be worse. Could be 5:3:2 for an effective rate of 18 fps on a 60p channel. Yep, this is real cadence from an ATSC 3.0 stream. These bastards try to save on bitrate with any means possible. Then again, why not 4:3:3, for example? It would be a tad smoother. Still crap, though.

[Brad]

A pulldown sequence with no odd numbers means there's no interlacing. "2:2:2:4" isn't really a thing

Why is it not a thing?

I don't mean you're describing it wrong; I mean it isn't a thing that any broadcaster, etc. is doing intentionally. It's purely a consequence of incompetence.

[manono]

Somewhat related: does anyone know, what software did hometheaterhifi use for listing flags, like here for the Beautiful Mind film?

No, but you can get similar information by opening a D2V file made by DGIndex using Notepad or similar. With soft pulldown it'll go 0 1 2 3. When it drops to video it'll go 0 0 0 0 or 2 2 2 2.