25p to 60i (using pulldown technique)

[manono]

Does it mean Sharc was wrong ?

To tell you the truth, I was puzzled by that, too, and figured he was referencing something earlier in the thread and I didn't feel like going back through the whole thing to figure it out. I also figured he was speaking of something else and we're not contradicting each other. In any event, any 60 (59.94) Hz display will have to output the video at either 59.94 fields per second or 59.94 frames per second. That's what pulldown, either hard or soft, is all about.

[Music Fan]

Thus it means players can't make IVTC with decimation with this kind of video.
But I still have doubts because, by comparison, 3:2 pulldown flag is supposed to create 29,97 fps, but some players can make IVTC and output 23.976 hz (when resolution is set on 1080p). So the flag is used but the reverse operation is done just after, inside the player, as if there was no flag.
Why wouldn't it be possible too with 25p flagged as 29,97 fps ?

[poisondeathray]

Thus it means players can't make IVTC with decimation with this kind of video.
But I still have doubts because, by comparison, 3:2 pulldown flag is supposed to create 29,97 fps, but some players can make IVTC and output 23.976 hz (when resolution is set on 1080p). So the flag is used but the reverse operation is done just after, inside the player, as if there was no flag.
Why wouldn't it be possible too with 25p flagged as 29,97 fps ?


Cadence reading setups that have 3:2:3:2:2 support can read it. 100% for sure.

When you look at hardware reviews, they test varous cadence tests. 3:2:3:2:2 is one of the (fairly) common patterns tested. Most newer players in the medium to more expensive range can. Cheapo models usually cannot. "FAIL" means they are deinterlaced (jaggies)

But you should understand that over here, 99.99% of the TV's run at 60Hz. Even a successfully IVTCed 3:2 "24p" will have judder because you 're seeing frame repeats. Even on a native 24p Blu-ray (no pulldown), you will see 3:2 repeats because the display is running at 60. Only those "24p" capable TV's that have a refesh rate of an integer multiple of 24 (e.g. 120hz, 240) can display them properly without any judder.

So you will see the same thing pattern wise. 3:2:3:2:2 on a 60Hz display, whether or not you have a smart IVTC or dumb deinterlacing setup. The difference is the picture quality

[Sharc]

....Why wouldn't it be possible too with 25p flagged as 29,97 fps ?
It's possible provided that the player can handle the 2:1:1:1:1 cadence. You would have to watch the "inverse telecined" stream on a PAL TV (50Hz) however to enjoy it judder-free.
Strictly, the 25<->30 (29.97) fps conversion with the 2:1:1:1:1 pattern is not Telecining or IVTC respectively. It's pure pulldown and decimation respectively. Telecining always involves fields and interlacing.

[manono]

...by comparison, 3:2 pulldown flag is supposed to create 29,97 fps, but some players can make IVTC and output 23.976 hz
Sort of. Flag readers (the vast majority of the NTSC DVD players out there) can make progressive 23.976fps for soft telecined material. Cadence readers can both do that as well as perform an IVTC on-the-fly, so to speak - return hard telecined DVDs to progressive 23.976fps. But, if it's a 60Hz progressive display (a nice 60" LED television, for example) that progressive 23.976fps video still has to be displayed in the 3232323232 sequence. As pdr says, it's not being deinterlaced so the picture quality is better, but it still has that 3-2 judder (is that the correct word?). The 120 and 240 Hz displays can play it smoothly, though,

[Sharc]

.... but it still has that 3-2 judder (is that the correct word?).
I think so. In the context of framerate conversion and pulldown it is the commonly used term for unequal display durations of individual frames on the screen ("pulldown-judder").
There are other sources of judder of course, e.g. due to motion and camera sweeps which are not synchronized with the TV playback framerate whatever this is. Modern TV's do a lot to mitigate these effects (e.g. by interpolating), more or less successfully.

[Music Fan]

It's possible provided that the player can handle the 2:1:1:1:1 cadence. You would have to watch the "inverse telecined" stream on a PAL TV (50Hz) however to enjoy it judder-free.
Of course, I'm not talking about 60hz only TV's. European TV's support 50 and 60hz, and also very often 24 hz (which is generally displayed in 96 hz, with or without motion compensation).
If a european player connected to a european TV (both supporting 50 and 60hz) can't make decimation with 25p flagged as 29.97, it's perhaps due to a choice (to stay in the Ntsc logic and display only 23.976 or 29.97 fps when the flag is 29.97), not a technical limit (admitting the player can make decimation).

@ poisondeathray and manono : I understand, but that does not really answer my question about the possibility to display in 50hz on a 50/60hz screen this kind of dvd (25p flagged as 29.97).

[Sharc]

.....
@ poisondeathray and manono : I understand, but that does not really answer my question about the possibility to display in 50hz on a 50/60hz screen this kind of dvd (25p flagged as 29.97).
Not sure whether I interpret your question correctly, but:
- by "25p source flagged as 29.97" you mean soft-pulldown i.e. every 5th frames is flagged to be pulled down (repeated) by the player at playback time. The original 25p stream is otherwise untouched, the pulldown is done at playback time by the player. Correct?
- A 50Hz (PAL) system will ignore the pulldown flag and play every 25p frame twice, or in a multiple of 50Hz depending on TV with/without "smart" motion interpolation.
- A 60Hz system will normally play it at 60 (59,94) Hz in a 4:2:2:2:2 pattern with judder. Smart systems may apply some interpolation to smoothen the playback
- A multi-norm system should recognize the pulldown flag and the 2:1:1:1:1 cadence, ignore it and display it in 50 Hz (or a multiple 100, 200 ....thereof).
I think these are the options, no?

[Music Fan]

Yes, but I wonder if a 50hz or multi-norm system can really ignore the flag (or use it then make the reverse operation because I believe the flag can't be ignored) and display this kind of dvd in 50hz.
But I'm not talking only about 29.97p but also 29.97i (what is usually called 60i). Thus whatever 3:2:3:2:2 or 2:1:1:1:1, if the video is in 25p and flagged with DGpulldown (to be displayed in 29.97 fps), the ideal framerate to display it is 50hz (50p). But that's maybe not possible with standalone players. I'm gonna try.

[Sharc]

Looking forward to your test results .... :)

[Music Fan]

Here are the results : I encoded in 720*480 @ 25p, added pulldown with DGpulldown (I checked the line 25 --> 29.97) and made a dvd with Muxman.
Tested on 3 players all set on 1080p : none displayed 50 hz !
Oppo and Pioneer output 60p, Sony outputs 24p !?

edit : if I disable 24p in Sony's setup (no instead of auto, there is not yes or force), it outputs 60p.
I can force 50p in Oppo's setup (which calls it PAL) ; that's probably the best way to look this kind of dvd but curiously I didn't notice difference compared to 60p output (when Multi-system is selected in Oppo's setup).

[Sharc]

Interesting. Not much smartness then.
It seems like the Sony applies standard IVTC with decimation.

[Music Fan]

I don't know what it does but 24p has no sense, that should be 50 or 60p.
I edited my previous post to add some informations.

[Sharc]

Hmmm.... Isn't your 25p source a speed-up 24 film?
Maybe the Sony is smarter than we thought... ;-)

At the end, the answer may again be: "It depends on the player".

[huhn]

Here are the results : I encoded in 720*480 @ 25p, added pulldown with DGpulldown (I checked the line 25 --> 29.97) and made a dvd with Muxman.
Tested on 3 players all set on 1080p : none displayed 50 hz !
Oppo and Pioneer output 60p, Sony outputs 24p !?

edit : if I disable 24p in Sony's setup (no instead of auto, there is not yes or force), it outputs 60p.
I can force 50p in Oppo's setup (which calls it PAL) ; that's probably the best way to look this kind of dvd but curiously I didn't notice difference compared to 60p output (when Multi-system is selected in Oppo's setup).

is this like soft telecine?

broadcast doesn't support this AFAIK.

[Music Fan]

Hmmm.... Isn't your 25p source a speed-up 24 film?
Yes.

Maybe the Sony is smarter than we thought... ;-)
No because it was encoded in 25p (with assumefps(25) in the script) and nothing can tell the player that the orignal speed was 24p, even the mpeg-2 encoder has no way to know it.
The script ouputs the same kind of signal than if I had de-interlaced 50i to 25p.
If I had done changefps(25) instead of assumefps(25), in this case ok, a frame would be duplicated and the player could detect it, but it's not what I did.

[Music Fan]

is this like soft telecine?
Yes, DGPulldown don't re-encode, it only add flags.

[Sharc]

Yes.


No because it was encoded in 25p (with assumefps(25) in the script) and nothing can tell the player that the orignal speed was 24p, even the mpeg-2 encoder has no way to know it.
The script ouputs the same kind of signal than if I had de-interlaced 50i to 25p.
If I had done changefps(25) instead of assumefps(25), in this case ok, a frame would be duplicated and the player could detect it, but it's not what I did.
I can only guess that the Sony player has seen the flags and then defaulted straight to 24, assuming that this is the most likely case. Just speculating.

[Music Fan]

I also believe this.

[Sharc]

Did you inspect the .ifo and .vob with e.g. ifoedit? Anything unusual? These contain also instructions for the playback.

Edit:
Just as a sidenote: When I do a 2:1:1:1:1 hard-pulldown (Changefps(30)) in avisynth and decimate it with tdecimate() without specifying the framerate explicitly it also defaults to 24 fps. In order to get 25fps output I need to specify tdecimate(mode=2,rate=25), or fdecimate(rate=25).

[TheSkiller]

Using interlaced chroma sampling is exactly the same as not doing anything special to the chroma, unless we're talking about one of those formats with oddball interlaced chroma sampling. (DV is one of them, I think?)It's not the same, I am not talking about chroma placement (like DV vs MPEG2). It's about how the chroma was downsampled from (usually) 4:2:2 to 4:2:0 for DVD encoding. I'm sure you know this can be done in two ways

ConvertToYV12(interlaced=false)
and
ConvertToYV12(interlaced=true).

For a progressive video, one would use interlaced=false, right? The problem is, if you encode (or rather flag) the video interlaced the player might do something like this to upsample the chroma for playback:

ConvertToYUY2(interlaced=true)

Now that will result in the CUE (chroma upsampling error) if you used ConvertToYV12(interlaced=false) rather than ConvertToYV12(interlaced=true) for downsampling to 4:2:0.



(Also, I just did a search on CUE and the first few results were from like 2001. How is this still a concern in 2015?)At that time it was a concern because many hardware players were simply lacking a progressive chroma upsampling method. They used an interlaced one, no matter what.

It is potentially still a concern today in the case described above, but this time it's not the hardware's fault but user error.


CUE (down: p; up: i)
https://picload.org/infonail/pdodgdp/cue.jpg (https://picload.org/image/pdodgdp/cue.png)

Correct (down: p; up: p)
https://picload.org/infonail/pdodgcp/correct.jpg (https://picload.org/image/pdodgcp/correct.png)

Also correct (down: i; up: i)
https://picload.org/infonail/pdodgwo/all_interlaced.jpg (https://picload.org/image/pdodgwo/all_interlaced.png)

[colours]

It's not the same, I am not talking about chroma placement (like DV vs MPEG2).

Neither was I—I was pointing out that if we assume that we can just Weave() together fields and have correct chroma siting (this is true of the "standard" 4:2:0 interlaced chroma siting), then there's nothing special that needs to be done if we can assume the playback setup isn't completely broken.

Okay, that's actually a huge bunch of "if"s chained together and we all know playback setups can be broken in very creative ways, but honestly I think the solution should be to fix them (or at least get people to stop buying shitty DVD players etc.) rather than messing with things on the mastering side.

It's about how the chroma was downsampled from (usually) 4:2:2 to 4:2:0 for DVD encoding. [snip] Now that will result in the CUE (chroma upsampling error) if you used ConvertToYV12(interlaced=false) rather than ConvertToYV12(interlaced=true) for downsampling to 4:2:0.

My ten minutes of Googling led me to believe that the term "CUE" is meant specifically for the problem of upscaling the chroma fields with the wrong offsets, not for a mismatch of interlaced/progressive resampling. The latter's not as big a problem and can be worked around on the mastering end just by using a low-pass filter prior to interlacing (or widening the support of the downscaling filter when converting from 4:2:2/4:4:4/RGB to 4:2:0).

A playback setup is not required to be able to detect and reverse pulldown, so this application of a low-pass filter is the only way to guarantee that the chroma will correctly render on playback, regardless of whether the content is really interlaced (50/60 temporally distinct fields) or pulled-down. In practice it's probably safe to assume some sort of field matching will be used and that progressive chroma upsampling is done after all the deinterlacing and field matching nonsense, so you could just ignore this issue entirely. If it still uses interlaced chroma upsampling after deinterlacing or field matching, then that obviously falls under the category of "your setup is broken so we won't support it".

[Music Fan]

And how to apply a low-pass filter with Avisynth ?

[kolak]

blur(1).sharpen(0.3).blur(1).sharpen(0.3)

[TheSkiller]

Neither was I—I was pointing out that if we assume that we can just Weave() together fields and have correct chroma siting (this is true of the "standard" 4:2:0 interlaced chroma siting), then there's nothing special that needs to be done if we can assume the playback setup isn't completely broken.I agree, nothing needs to be done.


In practice it's probably safe to assume some sort of field matching will be used and that progressive chroma upsampling is done after all the deinterlacing and field matching nonsense, so you could just ignore this issue entirely. If it still uses interlaced chroma upsampling after deinterlacing or field matching, then that obviously falls under the category of "your setup is broken so we won't support it".Yes, but if you are using a traditional plain ol DVD-player without any progressive scan, just Scart, the player wouldn't do anything more sophisticated than ConvertToYUY2/RGB(interlaced=true) whenever the stream is flagged as interlaced.
Now my point is, the player is not doing anything wrong there, really. Whenever you put something non-interlaced on a DVD and yet flag it interlaced, you should downsample the chroma to 4:2:0 the "interlaced way". I'm rather sure in the end that's probably almost the same as applying a vertical low-pass on the chroma, so we kind of agree that something needs to be done.

[Music Fan]

blur(1).sharpen(0.3).blur(1).sharpen(0.3)
Ok, and why does it act as low-pass ?

[kolak]

Ask Didee, it's his idea and it works.

[Sharc]

blur() reduces details and sharp transitions by removing high frequencies = lowpass filtering

You could also downscale (and then upscale) the picture. Scaling applies a lowpass filter in order to prevent aliasing.

[colours]

Whenever you put something non-interlaced on a DVD and yet flag it interlaced, you should downsample the chroma to 4:2:0 the "interlaced way". I'm rather sure in the end that's probably almost the same as applying a vertical low-pass on the chroma, so we kind of agree that something needs to be done.

I'm not entirely convinced by this, actually. Going by the very same pictures you uploaded, it's clear that even with the stronger colour fringes, the p-down/i-up pic looks less bad than the i-down/i-up pic.

Let's disregard the idea of using a low-pass filter for the time being. By using interlaced downscaling, there would be two steps causing aliasing in the whole chain (both downscaling and upscaling), whereas with progressive downscaling, the only step that causes aliasing is the interlaced upscaling step. There's no obvious reason to believe that the aliasing caused by interlaced upscaling and by interlaced downscaling should magically cancel out, so progressive downscaling should be preferred.

With a low-pass filter applied, all of this becomes moot if we use a sinc filter (zero aliasing in either step!), but since we don't actually use sinc filters for low-pass filtering or resampling (lolringing + lolslow), to minimise aliasing we should still use progressive downscaling. (Also, we can combine the resampling and low-pass filtering steps into one by adjusting Dither_resize16's fv parameter.)

I think there might be a better way of downscaling than either normal progressive downscaling or interlaced downscaling if we know that it'll be interlace-upscaled later on, but I'm a bit too busy to work out the details at the moment.

[TheSkiller]

I'm not entirely convinced by this, actually. Going by the very same pictures you uploaded, it's clear that even with the stronger colour fringes, the p-down/i-up pic looks less bad than the i-down/i-up pic.OK, those lines I drew in Paint aren't very good at showing the aliasing. So I re-did the test with this chroma zone plate.


down: p; up: i
https://picload.org/infonail/pdllpir/p-i.jpg (https://picload.org/image/pdllpir/p-i.png)

down: p; up: p
https://picload.org/infonail/pdllpcw/p-p.jpg (https://picload.org/image/pdllpcw/p-p.png)

down: i; up: i
https://picload.org/infonail/pdllppc/i-i.jpg (https://picload.org/image/pdllppc/i-i.png)