NTSC : progressive or interlaced

[mathmax]

Hello

In order to edit a video, I needed to deinterlace the file.
Now, I would like to make a NTSC DVD, and I wonder if it's needed to have an interlaced file.

I know that LCD TVs usually perform a simple bob in order display progressive frames on the screen. But if it applies such a filter on my video which is already deinterlaced, I will lose half of each frame .. and I guess lose quality.

I thought maybe there is a way to inform the player that the file is already progressive and doesn't need any additional bob filter to be displayed.

Could anyone bring his lights on this issue? Is it recommended (or required) to export NTSC interlaced? How can I inform that the file is already progressive and doesn't need any additional filter?

thanks

[edDV]

Hello

In order to edit a video, I needed to deinterlace the file.

No you don't. There are many editors that handle interlace video.

Now, I would like to make a NTSC DVD, and I wonder if it's needed to have an interlaced file.

See "What is DVD?" above left.

First "NTSC" formally means analog video but after digital capture to Y, Cb, Cr digital components, the term is often applied to 720x480 or other resolutions at 29.97 frame rate. An "NTSC" digital file is normally assumed to be interlace.

An "NTSC" DVD is normally 29.97 fps (54.94 fields per second) interlace. An alternative so called "progressive NTSC" DVD has a base 23.976 frame rate but stores the progressive frames as two fields with flags that tell the player to apply "pull down" (also called "telecine") to output at 29.97 fps rate as interlace video. A "progressive DVD player" offers a second option to output as progressive frames. The 23.976 fps progressive frames are frame repeated 2x then 3x at a 59.94 fps rate or 232323.

((2x 23.976)+(3x 23.976))/2 = (47.952 + 71.928)/2 = 59.94

Progressive output is only possible on analog component or HDMI outputs.

I know that LCD TVs usually perform a simple bob in order display progressive frames on the screen. But if it applies such a filter on my video which is already deinterlaced, I will lose half of each frame .. and I guess lose quality.

No. A typical LCD or plasma TV will accept 29.97 interlace video, then perform a bob deinterlace (motion adaptive or not*) to 59.94 frames per second progressive. Then the frames are upscaled to screen resolution and displayed. If the 29.97 fps interlace video input was progressive sourced (i.e. both fields were sampled at the same point in time), then the bob deinterlace results in a 2x 2x progressive frame repeat. In other words, 29.97p interlaced to 59.94 fields per second gets displayed as frame doubled 29.97p at a 59.94p rate.

Most current HDTV sets have so called "cinema processing" or telecine detection. The incoming video is analyzed for telecine field repeat sequence (aka cadence). When a telecine field sequence is detected, the video is inverse telecined back to 23.976p fps, then frame repeated 2x then 3x to 59.94 fps rate.

Progressive DVD player 59.94p gets displayed directly.

Note: I'm skipping over 120/240/480 Hz HDTV sets for this discussion.

I thought maybe there is a way to inform the player that the file is already progressive and doesn't need any additional bob filter to be displayed.

29.97p is a bastard frame rate that doesn't occur in a normal analog NTSC world. But when processed as 29.97 interlace (e.g. to 480i DVD) it will sneak through and get bobbed to 59.94 and displayed. The only down side is motion samples will still be at 29.97 rate so the video will look more stepped (half motion sampled) vs true interlace.

Could anyone bring his lights on this issue? Is it recommended (or required) to export NTSC interlaced? How can I inform that the file is already progressive and doesn't need any additional filter?

For DVD or analog recording/broadcast one must fake 29.97p as interlace. There are no header flags or other ways to identify 29.97p.

For ATSC or Blu-Ray, there are header flags for 29.97p so interlace conversion is not required.

Same goes for h.264. 29.97p (30p) is a supported frame rate.


* a simple bob deinterlace has flicker/wobble artifacts during low motion or stills. A motion adaptive bob deinterlace switches to single field (29.97p x2) mode when low motion is detected. A simple motion adaptive bobber applies field mode to the full frame. A sophisticated motion adaptive bobber will selectively apply field rate to low motion areas of the frame.

[mathmax]

Thank you very much for your detailed answer

No you don't. There are many editors that handle interlace video.

yes for some reasons I do need to deinterlace. I don't need special editor.. I use avisynth. I could process each field separatly, but the result would not be as good.

See "What is DVD?" above left.

I read but I don't see where they speak about interlacing.. you do however explained well the issue.

An "NTSC" DVD is normally 29.97 fps (54.94 fields per second) interlace. An alternative so called "progressive NTSC" DVD has a base 23.976 frame rate but stores the progressive frames as two fields with flags that tell the player to apply "pull down" (also called "telecine") to output at 29.97 fps rate as interlace video.

so it doesn't change the issue, cause when the player will invert the telecine, it will assume that the 29.97 fps video is interlaced.. and then apply a bob deinterlacer to display on screen, right?

No. A typical LCD or plasma TV will accept 29.97 interlace video, then perform a bob deinterlace (motion adaptive or not*) to 59.94 frames per second progressive. Then the frames are upscaled to screen resolution and displayed. If the 29.97 fps interlace video input was progressive sourced (i.e. both fields were sampled at the same point in time), then the bob deinterlace results in a 2x 2x progressive frame repeat. In other words, 29.97p interlaced to 59.94 fields per second gets displayed as frame doubled 29.97p at a 59.94p rate.

that's what I meant by losing half of the fames.. basically a frame will be spited in two "fields", each field will be stretched to 480px and then the player will either 59.94fps or 29.97fps. But the final framerate doesn't really matter for me, I just don't like the idea to drop half of a frame whereas it's already deinterlaced and could be shown directly on screen without any filter and resizing.

* a simple bob deinterlace has flicker/wobble artifacts during low motion or stills. A motion adaptive bob deinterlace switches to single field (29.97p x2) mode when low motion is detected. A simple motion adaptive bobber applies field mode to the full frame. A sophisticated motion adaptive bobber will selectively apply field rate to low motion areas of the frame.

not sure to understand what you mean by field mode/field rate... you mean both fields will be used in the areas with low motion? That would mean... in the case of a progressive source, field mode will be used for all the video, since there is no motion between the 2 fields already deinterlaced. So my video will be displayed at 59,94fps and the frames will be untouched, each of them being displayed 2 times. That will look exactly like a 29.97p played on computer, right? In that case, I don't mind so much it's like a bastard format.. as long as the player plays my video without any additional filter...

[edDV]

Back to the original goal. I think there is something basic that you are missing.

Hello

In order to edit a video, I needed to deinterlace the file.
Now, I would like to make a NTSC DVD, and I wonder if it's needed to have an interlaced file...
....

I am assuming your source file is 720x480i, 29.97 interlace.

For now lets ignore any need to use specific AVIsynth restoration filters.

If you Bob deinterlace to 59.94 fps what you are doing is vertically interpolating each 720x240 field to 720x480. Such interpolation isn't perfect. That is why low motion or still frame playback at 59.94 can show flicker due to field one vs field two interpolation errors.

Now you want to make a DVD. DVD by definition supports only 480i @29.97 (59.94 fields per sec) or 480p @ 23.976 (burned as 47.952 fields per sec). Since your source is 29.97 fps, you will need to use the first option.

In order to convert your bobbed 720x480p 59.94 fps for DVD, you need to re-interlace each interpolated frame and toss half the fields to get back to 59.94 fields per sec. This conversion 480i 29.97 to 480p 59.94 and back to 480i 29.97 is lossy and unnecessary. You should use the original 480i 29.97 source for DVD.

If editing is required, this can be done interlace mostly without loss depending on video format. GOP based MPeg source can be losslessly edited at the I frame. "Smart render" editors like Video Redo or Vegas can edit inside the GOP with re-encode only for the affected GOP.

I'll let you consider the above and return this evening to address your other questions.

[manono]

I could process each field separatly, but the result would not be as good.

Nonsense, not if you do it right (bob->filter->reinterlace). That is, bobbing before filtering gives you slightly better quality output than does separating the fields followed by filtering.

so it doesn't change the issue, cause when the player will invert the telecine, it will assume that the 29.97 fps video is interlaced.. and then apply a bob deinterlacer to display on screen, right?

The majority of the players out there are pure flag readers. They will deinterlace pure interlace and hard telecine both. A cadence reader will do an on-the-fly 'IVTC' of hard telecine. Both kinds will handle soft telecine well, with no deinterlacing at all. For progressive displays they'll output the full and original frames, (if the player is progressive scan).

[mathmax]

I am assuming your source file is 720x480i, 29.97 interlace.

right

Now you want to make a DVD. DVD by definition supports only 480i @29.97 (59.94 fields per sec) or 480p @ 23.976 (burned as 47.952 fields per sec). Since your source is 29.97 fps, you will need to use the first option.

on another thread someone showed me a sample of a DVD. It's 480p @29.97:
http://www.mediafire.com/download.php?zau76d1lamqedq7
It seems to be like my file after editing.. how would such file will be played on screen? Will the player assume the file is interlaced and apply a bob filter, which means split each frame, interpolate the fields to 480px and double the framerate? Or will it recognize that the file is already progressive and display it on screen as it is, without additional deinterlacing filter?

In order to convert your bobbed 720x480p 59.94 fps for DVD, you need to re-interlace each interpolated frame and toss half the fields to get back to 59.94 fields per sec. This conversion 480i 29.97 to 480p 59.94 and back to 480i 29.97 is lossy and unnecessary. You should use the original 480i 29.97 source for DVD.

ok.. but my avisynth script needs to work on deinterlaced frames. I could perform on each field separately... manono advised to apply a bob before to apply my script, but you don't recommend to convert 480p 59.94 and back to 480i 29.9... So what can I do?
So far I deinterlace using YadifMod(edeint=nnedi3()), so it doesn't change the framerate.. then I can apply my script on the progressive frames.

[manono]

on another thread someone showed me a sample of a DVD. It's 480p @29.97:
http://www.mediafire.com/download.php?zau76d1lamqedq7
It seems to be like my file after editing.. how would such file will be played on screen? Will the player assume the file is interlaced and apply a bob filter, which means split each frame, interpolate the fields to 480px and double the framerate?

Both samples are encoded as progressive. Although I see progressive 29.97fps content usually encoded as interlaced on retail DVDs, because yours is encoded as progressive, both flag reading and cadence reading progressive scan players will output whole frames to the progressive display. No interpolation or bobbing. Just each frame played twice for a 59.94fps progressive display, such as an LCD Hi-Def TV set.

DVD is an interlaced medium, as edDV says. But in the case of progressive scan players each field of your progressively encoded 29.97fps samples is intercepted and rejoined with its partner to make a complete frame again before final output.

manono advised to apply a bob before to apply my script, but you don't recommend to convert 480p 59.94 and back to 480i 29.9... So what can I do?

If you do something like this:

AssumeTFF() #if TFF
Bob with your favorite bobber (not QTGMC, though)
Filter
SeparateFields()
SelectEvery(4,0,3)
Weave()

then you'll be OK.

The reason I say not to use QTGMC for something like this is because it messes with both fields where with normal bobbers one field of the resulting frame is untouched and the other is interpolated. Then when the fields are separated again and reinterlaced, you'll get back the original one (original except for whatever the filtering did to it). Plus QTGMC is usually unnecessarily slow. As I understand it, even a crappy bobber such as Bob() is OK, although I'll use Yadif which is almost as fast.

[edDV]

Thank you very much for your detailed answer

No you don't. There are many editors that handle interlace video.

yes for some reasons I do need to deinterlace. I don't need special editor.. I use avisynth. I could process each field separatly, but the result would not be as good.

See "What is DVD?" above left.

I read but I don't see where they speak about interlacing.. you do however explained well the issue.

An "NTSC" DVD is normally 29.97 fps (54.94 fields per second) interlace. An alternative so called "progressive NTSC" DVD has a base 23.976 frame rate but stores the progressive frames as two fields with flags that tell the player to apply "pull down" (also called "telecine") to output at 29.97 fps rate as interlace video.

so it doesn't change the issue, cause when the player will invert the telecine, it will assume that the 29.97 fps video is interlaced.. and then apply a bob deinterlacer to display on screen, right?

No. A typical LCD or plasma TV will accept 29.97 interlace video, then perform a bob deinterlace (motion adaptive or not*) to 59.94 frames per second progressive. Then the frames are upscaled to screen resolution and displayed. If the 29.97 fps interlace video input was progressive sourced (i.e. both fields were sampled at the same point in time), then the bob deinterlace results in a 2x 2x progressive frame repeat. In other words, 29.97p interlaced to 59.94 fields per second gets displayed as frame doubled 29.97p at a 59.94p rate.

that's what I meant by losing half of the frames.. basically a frame will be spited in two "fields", each field will be stretched to 480px and then the player will either 59.94fps or 29.97fps. But the final framerate doesn't really matter for me, I just don't like the idea to drop half of a frame whereas it's already deinterlaced and could be shown directly on screen without any filter and resizing.

The main "What is DVD?" point is DVD players and TV sets will process the video with the expectation that it fits into one of three standard categories.

1. 29.97 interlace is the default expectation. A typical progressive DVD player will bob the fields to 59.94 frames per sec (using line doubling or missing line interpolation). The expectation is that each field represents a 1/59.94 sample in time, so the resulting 59.94 frames represent unique time samples over a second.

2. Soft telecine header flags tell the progressive player to weave progressive frames from stored fields and repeat them in a 3x 2x sequence to 59.94 fps over analog component or HDMI.

3. Hard telecine detection triggers an inverse telecine and 3x 2x frame repeat to 59.94 fps.

To fit 29.97p video into this scheme, you must disguise it as interlace. The progressive frame is split into two fields for storage on the DVD. Ideally you would want the player to weave these fields back into a progressive frame and then 2x frame repeat to 59.94 but that isn't what happens. Instead for simple players, each field is line doubled or interpolated from 720x240 to 720x480 to make a 59.94 fps output. The TV will see this as 59.94p and simply upscale. The resulting displayed frames will have errors resulting from the player's line interpolation. These errors are compounded in the upscale process. Some DVD players will have more sophisticated deinterlacers that analyze MPeg motion vectors and apply a weave, blend or bob depending on the amount of field to field motion. These will produce better results from 29.97p because field to field motion will be zero.

A computer DVD player will go through a similar process in "auto" mode but it is possible for the user to force deinterlace into a weave or blend mode when playing 29.97p source.

[edDV]

* a simple bob deinterlace has flicker/wobble artifacts during low motion or stills. A motion adaptive bob deinterlace switches to single field (29.97p x2) mode when low motion is detected. A simple motion adaptive bobber applies field mode to the full frame. A sophisticated motion adaptive bobber will selectively apply field rate to low motion areas of the frame.

not sure to understand what you mean by field mode/field rate... you mean both fields will be used in the areas with low motion? That would mean... in the case of a progressive source, field mode will be used for all the video, since there is no motion between the 2 fields already deinterlaced. So my video will be displayed at 59,94fps and the frames will be untouched, each of them being displayed 2 times. That will look exactly like a 29.97p played on computer, right? In that case, I don't mind so much it's like a bastard format.. as long as the player plays my video without any additional filter...

It depends on the sophistication of the player's deinterlacer. As said above, the players expectation is normal interlace so it will treat each field as a separate time sample. The next level of processing is motion analysis. If the deinterlacer is in communication with the MPeg2 decoder, it can get an indication of frame to frame or field to field motion from the MPeg motion vectors. The deinterlacer can then switch modes from weave to blend to bob based on the amount of motion. This type of processing is becoming typical in DVD player and HDTV chip sets. The next level of processing stores motion data over several frames allowing deinterlace optimization at the object level within the frame. This level is currently too expensive for parts and royalties to be widely implementred.

So the fate of 29.97p depends on whether motion is detected at the frame or field level. Since the player isn't expecting 29.97p, it may not be looking for it.

The better chipsets include telecine cadence detection. It would be possible to detect 29.97p as a 2:2 sequence similar to the way PAL players do. PAL players apply a weave when the 2:2 pattern is detected.

The best way to sum this up is individual players need testing to see how 29.97p is handled.


PS: Manono stated above that the 2:2 cadence is detected on modern progressive players. That solves your problem.

[edDV]

In order to convert your bobbed 720x480p 59.94 fps for DVD, you need to re-interlace each interpolated frame and toss half the fields to get back to 59.94 fields per sec. This conversion 480i 29.97 to 480p 59.94 and back to 480i 29.97 is lossy and unnecessary. You should use the original 480i 29.97 source for DVD.

ok.. but my avisynth script needs to work on deinterlaced frames. I could perform on each field separately... manono advised to apply a bob before to apply my script, but you don't recommend to convert 480p 59.94 and back to 480i 29.9... So what can I do?
So far I deinterlace using YadifMod(edeint=nnedi3()), so it doesn't change the framerate.. then I can apply my script on the progressive frames.

In my statement I was setting aside the filter issue. It depends what types of filtering needs to be done. Many filters can be applied to interlace video, some require progressive conversion first.

Since DVD can't accommodate 59.94p source, most people will deinterlace to 29.97p, filter then re-interlace for DVD. The downside is half the temporal detail is lost.

Unfortunately Blu-Ray lacks 59.94p support at 720x480 or 1920x1080 as well. Only 1280x720 is supported for 59.94p.

[manono]

To fit 29.97p video into this scheme, you must disguise it as interlace.

Meaning progressive content encoded as interlaced, as is most common with PAL DVDs and 29.97fps progressive in NTSC DVDs? If so, then yes, in most cases it gets bobbed one way or another and only decent cadence reading players have 2:2 pulldown detection to weave the two fields of each frame back together.

PS: Manono stated above that the 2:2 cadence is detected on modern progressive players. That solves your problem.

But manono was writing specifically about the two samples where 29.97fps progressive content was encoded as progressive. For 29.97fps progressive content encoded as interlaced, I defer to your explanation as to what's going on within most players.

[mathmax]

Thank you so much for taking the time to answer my questions and explain in detail.

It's strange that we still bother with the interlacing issue... as I understand it, it was made for CRT TVs... now they are almost not existing anymore and the players still need sophisticated algorithms to deal to a greater or lesser extend with the issue, trying to recreate a progressive image to the viewers.

I'm also surprised because most of the DVD we can buy on stores aren't interlaced. So almost all of them must be in the category "Soft telecine with header flags".. right? But apparently I'm not alone to produce 29,97p.. if I rely on the samples I posted in my previous post..

So.. the progressive scan players will be able to play my DVD without bob. That indeed solve my problem.. if they are widely implemented. I want to produce my DVD for several persons... but I don't know their devices.

manono, thank you for the trick. You said that with normal bobbers one field is untouched and the other is interpolated. Aren't they both resized to 480p? If I use separatefield(), I'll loose half of the resized fields... I think it's lossy to resize and then drop half of the lines, isn't it?

Since DVD can't accommodate 59.94p source, most people will deinterlace to 29.97p, filter then re-interlace for DVD. The downside is half the temporal detail is lost.

I guess you meant the contrary. DVD can't accommodate 29.97p source.. so the people will deinterlace to 59.94p, right?

For 29.97fps progressive content encoded as interlaced, I defer to your explanation as to what's going on within most players.

encoded as interlaced? You mean the info interlaced or progressive is stored in the encoded file? So I should set an option to specify my DVD is progressive?

[manono]

Yeah, but at the time the DVD standards were being developed (in the mid 90's?), interlaced CRTs were about the only game in town. DVDs were developed for interlaced displays

manono, thank you for the trick. You said that with normal bobbers one field is untouched and the other is interpolated. Aren't they both resized to 480p?

Isn't your source another 480 DVD? If so, when being bobbed one field of the new frame remains unchanged and the other is interpolated in one way or another. If you are resizing (like from PAL to NTSC, or even doing a little cropping and then resizing), then one field of each new frame created when being bobbed will be mostly unchanged and the other interpolated. Doing a SeparateFields (as opposed to bobbing) also keeps the fields unchanged before filtering, but the problem there is the chroma blurring that takes place (or something like that - ask jagabo or pdr or edDV why filtering the fields before reinterlacing is inferior to bobbing first).

So I should set an option to specify my DVD is progressive?

Both of your samples had progressive 29.97fps encoded as progressive. That's not normal. Ordinarily progressive 29.97fps content is encoded as interlaced. I've done that a few times myself but, again, almost all of the ones I've seen (there are a lot of talking head interview extras on DVDs these days using progressive 29.97fps cameras) have been encoded as interlaced. Also, neither of those samples was from a retail DVD or most likely you could have proven it for yourself.

[Gavino]

Doing a SeparateFields (as opposed to bobbing) also keeps the fields unchanged before filtering, but the problem there is the chroma blurring that takes place (or something like that - ask jagabo or pdr or edDV why filtering the fields before reinterlacing is inferior to bobbing first).

There's no chroma blurring involved. The issue is that odd and even fields have different vertical alignment.

Spatial filtering (with no temporal element) can be done without problems on separated fields. However, temporal filters will see data from a different line when looking at neighbouring fields. This can be countered by filtering odd and even fields independently, but then you are losing half the temporal resolution in this filtering. Bobbing produces the correct vertical alignment while retaining full temporal resolution.

[davexnet]

Interesting. If the 29.97 progressive clip is encoded in HCenc and "interlacing options" is set to "auto-detect",
the program will encode a progressive mpeg-2. Selecting "make DVD compliant" doesn't alter the setting.

At least in Hank's thinking, progressive 29.97 mpeg-2 is as it should be for progressive source.

[manono]

At least in Hank's thinking, progressive 29.97 mpeg-2 is as it should be for progressive source.

If fed a progressive source, even a PAL progressive source, won't HCEnc always choose to encode it as progressive? But that flies in the face of actual practice where something like 99% of movies on retail PAL DVD are encoded as interlaced, and almost all NTSC progressive 29.97fps sources are also encoded as interlaced. Of course, it makes more sense to encode a progressive source as progressive, and a larger number of players will play it back without deinterlacing it.

[edDV]

It's strange that we still bother with the interlacing issue... as I understand it, it was made for CRT TVs... now they are almost not existing anymore ...

Interlace is fundamental to broadcasting. When the DVD was developed in 1997-98, all broadcasting was analog NTSC/PAL/SECAM interlace as were the VCR, Laserdisc and home camcorders. Without interlace support, DVD recorders wouldn't have been possible.

What was impressive about DVD was provision for storing progressive source (film then) as frames (stored as fields) rather than as hard telecine. The immediate benefit was 20% greater storage capacity for movies and extras. But it also opened the possibility of a progressive DVD player and a progressive projector, the first home digital cinema. This combined with 5.1 AC3 or DTS sound marked the beginnings of home theater.

The first DVD players only outputted interlace. An interlace disc could be played by simply outputting fields in sequence. A progressive disc was read in telecine field sequence (A1A2, B1B2, B3C1, C2D1, D2D3). Storage as fields combined with maximum 1GB VOB files meant affordable players could be manufactured in quantity.

The progressive player was held for electronic and chip technology to catch up. The development of the progressive DVD player more or less paralleled ATSC/DVB HDTV. Early players were targeted to progressive projectors. The incorporation of 480p and 720p into the HDTV standard (at the insistence of ABC) meant future HDTV sets would need to support both interlace and progressive scan.

Zoom ahead 10 years, progressive DVD player electronics have now been reduced to a few chips and all HDTV sets are progressive scan capable.

... the players still need sophisticated algorithms to deal to a greater or lesser extend with the issue, trying to recreate a progressive image to the viewers.

I'm also surprised because most of the DVD we can buy on stores aren't interlaced. So almost all of them must be in the category "Soft telecine with header flags".. right? But apparently I'm not alone to produce 29,97p.. if I rely on the samples I posted in my previous post..

Interlace still dominates HD broadcast technology because it is more efficient to transmit than equivalent progressive. 59.94 motion samples per second (fields or frames) remains the live TV standard. 24p remains the film and digital cinema standard. 29.97p is generally not used in broadcasting. It is too jerky on the big screen. 29.97p is mostly used for computer and mobile device display where bandwidth is restricted. Improvements in h.264 compression means 59.94p at full 1920x1080 will soon be the standard for live broadcast and for home camcorders.

Commercial DVD and Blu-Ray discs come in interlace or progressive formats. Since most titles are movies or film shot TV series, progressive dominates the titles but live shot TV series (e.g SNL, sports or reality) and most PAL discs are interlace format.

[mathmax]

Isn't your source another 480 DVD? If so, when being bobbed one field of the new frame remains unchanged and the other is interpolated in one way or another. If you are resizing (like from PAL to NTSC, or even doing a little cropping and then resizing), then one field of each new frame created when being bobbed will be mostly unchanged and the other interpolated.

My source is indeed a 480p DVD. I meant each field is resized from 240 to 480 when being bobbed... so I was surprised when you said they were untouched, cause to resize 240->480 and then keep only half of the lines seems to be a lossy operation, isn't it?

Both of your samples had progressive 29.97fps encoded as progressive. That's not normal.[...]
Of course, it makes more sense to encode a progressive source as progressive, and a larger number of players will play it back without deinterlacing it.

how do you see they are encoded as progressive? In AVStoDVD, I didn't check the option "interlaced encoding", is that correct?

I'm a bit confused what you mean by "not normal". You mean it's not the actual practice, but it's better to encode a progressive source as progressive, right?

59.94 motion samples per second (fields or frames) remains the live TV standard. 24p remains the film and digital cinema standard. 29.97p is generally not used in broadcasting. It is too jerky on the big screen.

If 29.97p is too jerky but 24p is ok ? I thought 29,97fps is high enough to render a smooth image.. to tell the truth, I didn't even expect that 59.94fps would make a difference with 29,97p...

[edDV]

59.94 motion samples per second (fields or frames) remains the live TV standard. 24p remains the film and digital cinema standard. 29.97p is generally not used in broadcasting. It is too jerky on the big screen.

If 29.97p is too jerky but 24p is ok ? I thought 29,97fps is high enough to render a smooth image.. to tell the truth, I didn't even expect that 59.94fps would make a difference with 29,97p...

24p requires special camera lenses and camera technique. You can do the same with 30p.

But if you take 29.97i (59.94 fields per sec) and drop half the fields for 29.97p, it will be jerky indeed (half the motion samples). Do your own tests.

OK for talking heads from locked down camera but jerky for action or hand held camera.

[mathmax]

mmm... I'm not expert but when I deinterlaced to 29.97fps using yadif(), the video didn't seem to be jerky...

so you think I should bob deinterlace to 59.94p, apply my filter and then reinterlace as manono advised? I'm a bit confused cause you said it's lossy but he seems to say the contrary
Which bob would you advice in this case? Yafif(mode=1)? Is there any better?

[manono]

I meant each field is resized from 240 to 480 when being bobbed... so I was surprised when you said they were untouched, cause to resize 240->480 and then keep only half of the lines seems to be a lossy operation, isn't it?

In the new frame, one field is kept unchanged and the other is 'new'. In a simple bob (the avisynth 'bob'), the other field is very lossy. With better bobbers any static portions are the same in both fields. In motion portions, the quality is dependent on how good the motion estimation and what other filtering is done to create that new field. So, for a 'talking head' news broadcast where much of the frame is unchanged from frame to frame, bobbing it well can be very easily accomplished. Then when reinterlacing it after being filtered, you'll want to use that original field, the untouched one, except for whatever filtering was done to it.

how do you see they are encoded as progressive? In AVStoDVD, I didn't check the option "interlaced encoding", is that correct?

How do I tell? Easy. Use ReStream, Bitrate-Viewer, DGIndex or other programs. If you didn't check the interlaced box, then by definition it gets encoded as progressive - the progressive frame flag gets set. There are only two choices, interlaced or progressive, although it's not uncommon for the frame type to switch back and forth with some regularity.

I'm a bit confused what you mean by "not normal". You mean it's not the actual practice, but it's better to encode a progressive source as progressive, right?

Yes, it's not the actual practice, usually. Me, I've done it both ways.

mmm... I'm not expert but when I deinterlaced to 29.97fps using yadif(), the video didn't seem to be jerky...

It may not, especially if you're used to the jerkiness of 3:2 pulldown (as am I) and don't find it jerky. But surely you can understand that a video sampled 30 times a second will play less smoothly than one sampled 60 times a second. What's the ideal framerate to render a smooth image. Peter Jackson is filming The Hobbit at 48fps and many think a framerate around there is nice and smooth. Thomas Edison thought 46 fps was good, but the cost of film prevented him from using framerates that high all that often.

I'm a bit confused cause you said it's lossy but he seems to say the contrary

Avisynth has a number of interlace-aware filters, but not all of them, not by a long shot. It's not a good idea to blindly deinterlace to 29.97fps as you were suggesting at the very beginning of this thread. And that's what edDV was responding to originally, your claim that it was necessary to deinterlace an interlaced video to filter it. Deinterlacing (to 29.97fps progressive) is quite different from bobbing (to 59.94fps progressive) before then reinterlacing it.

[jagabo]

I thought 29,97fps is high enough to render a smooth image.. to tell the truth, I didn't even expect that 59.94fps would make a difference with 29,97p...

It's not. See this post and the video:
https://forum.videohelp.com/threads/307004-Best-framerate-conversion-%28eg-23-97-to-30-...=1#post1888926

[mathmax]

In the new frame, one field is kept unchanged and the other is 'new'. In a simple bob (the avisynth 'bob'), the other field is very lossy. With better bobbers any static portions are the same in both fields. In motion portions, the quality is dependent on how good the motion estimation and what other filtering is done to create that new field.

Sorry, I don't understand that.. when I look in avisynth, the bob filter seems to edit both fields.
I compare field by field:

    SeparateFields()
    lanczosresize(width, 480)

and

    bob() #or even yadif(mode=1)

in the second case, the two fields seem more blurry.

but at the beginning, I wasn't even mentioning this loss of quality. I was mentioning that the process - to upscale one field to 480p and then drop half of it - is lossy itself.

As I understand it, there are two steps where I lose quality:
- bobbing produces the correct vertical alignment but the interpolation also introduce some imprecision, blur or whatever
- bobbing resizes each field from 240p to 480p. Then speratefield() will indeed recover 240p but also lose half of the lines and maybe some details

I must misunderstand something.. I hope you can correct me if I'm wrong.

How do I tell? Easy. Use ReStream, Bitrate-Viewer, DGIndex or other programs.

mmm... here is a screenshot of Bitrate-Viewer. Which info says the source is progressive?

It's not a good idea to blindly deinterlace to 29.97fps as you were suggesting at the very beginning of this thread. And that's what edDV was responding to originally, your claim that it was necessary to deinterlace an interlaced video to filter it. Deinterlacing (to 29.97fps progressive) is quite different from bobbing (to 59.94fps progressive) before then reinterlacing it.

well.. in which case would you deinterlace to 29.97fps? I thought that deinterlacing to 29.97fps retain more details on each frame than bobbing.. is it not true?
Also, yadifmod(edeint=nnedi3()) gives me a better result than yadif().. but I don't know how to you yadifmod() to produce 59.94p.

jagabo, thank you for the video

[manono]

mmm... here is a screenshot of Bitrate-Viewer. Which info says the source is progressive?

Wrong Bitrate-Viewer. I mean the Tecoltd one. Isn't that the one you see when clicking on the link?

As for one field of a bobbed frame being indentical to the original one, I got that from Alex_ander and have never had reason to doubt it (or him). My guess is you're not doing it correctly. Take a frame. Separate the fields. Set them aside. Take the same original frame and use Bob on it. Separate the fields of the two resulting frames. One field from each bobbed frame should be identical to one of the fields from the original frame. If that makes any sense.

[mathmax]

Wrong Bitrate-Viewer. I mean the Tecoltd one. Isn't that the one you see when clicking on the link?

sorry for the confusion.. Now I tried to install the one you showed, but it doesn't work on win 7.
where can I see in dgindex?

As for one field of a bobbed frame being indentical to the original one, I got that from Alex_ander and have never had reason to doubt it (or him). My guess is you're not doing it correctly. Take a frame. Separate the fields. Set them aside. Take the same original frame and use Bob on it. Separate the fields of the two resulting frames. One field from each bobbed frame should be identical to one of the fields from the original frame. If that makes any sense.

so.. I compare:

SeparateFields()

field 1: http://img521.imageshack.us/img521/3774/separate0000.png
field 2: http://img862.imageshack.us/img862/9696/separate0001.png

AssumeTFF()
Bob()
SeparateFields()
SelectEvery(4,0,3)

field 1: http://img208.imageshack.us/img208/2356/bob0000.png
field 2: http://img820.imageshack.us/img820/3496/bob0001.png

the last ones are more blurry, both have lost quality.

[jagabo]

I agree that bob() doesn't work quite right. But Yadif does.

WhateverSource()
AssumeBFF()
v1=SeparateFields().Subtitle("sep")
v2=Yadif(mode=1, order=0).SeparateFields().SelectEvery(4,0,3).Subti tle("yadif")
Interleave(v1,v2)

[manono]

sorry for the confusion.. Now I tried to install the one you showed, but it doesn't work on win 7.

Oh, too bad. It's working on a computer of mine with Win7 32 bit. Are you using a 64 bit version?

Anyway, to confirm that both samples were encoded as progressive, open them in DGIndex and run the Preview. In the Information Panel you'll see that the Video Type is NTSC (29.97fps) and the Frame Type is Progressive.

[edDV]

mmm... I'm not expert but when I deinterlaced to 29.97fps using yadif(), the video didn't seem to be jerky...

so you think I should bob deinterlace to 59.94p, apply my filter and then reinterlace as manono advised? I'm a bit confused cause you said it's lossy but he seems to say the contrary
Which bob would you advice in this case? Yafif(mode=1)? Is there any better?

I see there has been post activity while I was gone but I wrote this when I got back and will post it then catch up.

The reason you normally use Yadif mode=1 is to convert a 59.94 fields/s interlace stream to 59.94 frames/s to display interlace on progressive screens.

You don't normally do this to go to DVD because DVD will only accept 59.94 fields/s or 23.976 frames/s.

You go to 59.94 frames/s because your filter requires it (for example a translation filter to tip/rotate video back into perspective). Lets use that example because it is graphic.

Input source is 720x480i in DV format. To apply this translation filter I must first convert incoming fields to progressive frames at 59.94 rate. This preserves all the motion samples. The translation filter remaps the pixels using interpolation to fit user xyz control requests.



If you preview play this file, the speaker's lip sync and motion detail is preserved at 59.94 rate. A sophisticated viewer may see some artifacting from Yadif field to frame conversion or from the translation filter. Most viewers will just see the effect.

Now we have a 59.94 frame per second version of the effect. That won't work for DVD so we must re-interlace to 480i. This is where it gets interesting. Re-interlacing will interpolate new fields matched to progressive frame pairs. The first frame is re-interpolated to 720x240 field one and the second is re-interpolated to 720x240 field two. This involves re-sampling horizontal lines alternately back to 720x480i.


The red lines indicate field rate re-sampling.

In this example the result conserves 59.94 motion samples to the 480i DVD. The speaker's lip sync and mouth movement detail are preserved. Further the frame translation movement motion detail is also at 59.94 rate (i.e. not stepping). Is there a down side? Sure there will be a certain amount of residual interpolation error from Yadif Mode=1 and from the translation filter. But the effect will appear smooth and most viewers won't notice interpolation errors.

If you did the same effect at 29.97p rate, the speakers mouth would move at half rate and the frame translation movement from full screen to the orientation shown would appear stepped.

There may be an argument for other filters using 29.97p processing or Jagabo may come in to tell us we are destroying the chroma with these translations.

I hope this helped show why maintaining 59.94 motion resolution is important.

[mathmax]

manono, yes I use window 7- 64 bit. But I could see the info in dgindex, thank you

I agree that bob() doesn't work quite right. But Yadif does.

Indeed, they perfectly match when I use Yadif(mode=1)
And I understand now what manono meant. On each interpolated field, one line over two is untouched an the others are interpolated.
So:

AssumeTFF()
yadif(mode=1)
SeparateFields()
SelectEvery(4,0,3)

is equivalent to SeparateFields()

but

AssumeTFF()
yadif(mode=1)
SeparateFields()
SelectEvery(4,1,2)

...is very lossy.

Now, if I compare Yadif(mode=1) and separatefields() + lanczosresize(width, 480), I realize that the resized field looks better than the interpolated one.

resized:
http://img341.imageshack.us/img341/3774/separate0000.png

interpolated:
http://img210.imageshack.us/img210/7062/yadif0000.png

Of course that doesn't take the motion into account.. and focus on the quality of a single frame. But that makes me wonder if it's ideal to apply spatial filters on interpolated fields... I risk to introduce errors on the untouched lines. And even if the interpolated field looks nice after I apply the filter, not sure it'll still look so nice after reinterlacing and bob() performed by the player... I mean, since half of the lines will be dropped and reinterpolated, the final interpolated frame might look different...

So.. maybe this technique is better for motion. But is it really better if you consider the quality of each frame?
yadifmod(edeint=nnedi3()) gives me a nice result on which I can apply my filters and I know they'll not be damaged by future interpolations. Of course I lose half of the framerate.. but since I don't notice my video is jerky, I wonder if it's really worth to conserves 59.94 motion samples...

Anyway, I read carefully your explanations edDV and surely there is a real advantage to keep 59.94. Since it's new for me, I might not judge well is the best choice for my DVD.. but I try to understand the issues involved by each solution. I'm very grateful you take the time to explain.

[manono]

I agree that bob() doesn't work quite right. But Yadif does.

Thanks for checking. I must have gotten garbled what Alex_ander said and maybe he was talking about Yadif and similar bobbers, and not the basic Bob().