Using 24p for AVS HD has come up as a topic of interest. I'm a little confused on what happens with HD DVD, and was hoping someone could set me straight.

It is my understanding from dr394 (Ron) that, in order to get 24p output from HD DVD, you must encode a 23.967 fps stream with pulldown flags, and that stream must be interlaced.

Now on to my question. We have a pattern that consists of alternating black and white, single pixel height, horizontal lines. When encoded as interlaced, one field is completely black and the other is completely white. What should be expected as this pattern is deinterlaced for display on a 1080p TV?

I'm inclined to think that a player outputting 24p will recognize the pulldown and use a film mode deinterlace to restore the original progressive frames, thereby preserving the pattern.

My concern is with HD DVD players that only output 1080i... could the target display mistake the 1080i59.97 as video instead of film and apply a video mode deinterlace? It seems as though this would produce a smeared gray image or something that flickers. The same question would apply to players that only output 1080p29.97, can we count on the player to deinterlace correctly?

EDIT: One more question. alluringreality has found using his XA2 that the current disc, which is 29.97 fps interlaced (no pulldown), displays this pattern correctly. I would have thought that the XA2 would have picked up the stream as video instead of film, and applied a video mode deinterlace to get 1080p59.94, which would lead to a smeared picture. Why doesn't this happen? Is 1080p59.94 not supported by current TV's, so the XA2 just puts the two fields back together even though they would be from a different time in a real video based clip?

Don't think I'll attempt parsing all that, but 30 frames per second is achieved with CRT-type displays when our eyes/brain merge two 1/60-sec TV fields (half-frames) from standard 1080/60i (30i) HDTV. Fixed-pixel displays use a video processor that creates 1080@60p with a buffer memory that temporarily stores a TV field and combines it with the second 1/60-sec TV field.

A HD disc outputting 1080/60i (30i) can act like a live 1080i station delivering sports programming, but most HD disc players providing 1080/60i output have movies encoded with 2-3 pulldown. That means 24p, stored as TV fields on the disc, becomes 1080/60i; 2-3 pulldown adds extra TV fields, along with a slight speed change, converting 24p into 60i. That's also how most 24p-based HD programming is delivered (OTA, cable, DBS).

Displays or video processors with inverse pulldown can discard the added TV fields and stitch together the original 24p frames by deinterlacing two TV fields per frame. Many sets, unfortunately, than add pulldown again, displaying 1080p60, and introducing potential judder instead of displaying 24p at even multiples such as 48, 72, or 120 Hz. A limited number of displays can accept 24p deinterlaced within HD-disc machines and output at 1080/24p. Displays--except for CRT models--must also be able to deinterlace 1080/60i into 1080p60.

AIUI, most HD hardware can handle both the fractional Hz rates (so-called NTSC offset, even though HD isn't NTSC) or whole-number Hz rates. The H/DTV ATSC format has 18 frame/raster combinations, but actually 36 if you factor in the 0.1% NTSC offset. Sure I left out something, but all these parameters really takes a book. Matter of fact, a good one is C. Poynton's 2003 Digital Video and HDTV.-- John

Don't think I'll attempt parsing all that, but 30 frames per second is achieved with CRT-type displays when our eyes/brain merge two 1/60-sec TV fields (half-frames) from standard 1080/60i (30i) HDTV. Fixed-pixel displays use a video processor that creates 1080@60p with a buffer memory that temporarily stores a TV field and combines it with the second 1/60-sec TV field.

Unless my math is wrong, or you repeat each frame twice for some reason, you can't get a 60p stream with that method. If you simply hold one field in a buffer and combine the next one with it, those two fields still only represent 1/60th of a second each, or 1/30th of a second for the entire frame. You would have to repeat each frame twice to get 60 fps.


A HD disc outputting 1080/60i (30i) can act like a live 1080i station delivering sports programming, but most HD disc players providing 1080/60i output have movies encoded with 2-3 pulldown. That means 24p, stored as TV fields on the disc, becomes 1080/60i; 2-3 pulldown adds extra TV fields, along with a slight speed change, converting 24p into 60i. That's also how most 24p-based HD programming is delivered (OTA, cable, DBS).

Displays or video processors with inverse pulldown can discard the added TV fields and stitch together the original 24p frames by deinterlacing two TV fields per frame. Many sets, unfortunately, than add pulldown again, displaying 1080p60, and introducing potential judder instead of displaying 24p at even multiples such as 48, 72, or 120 Hz. A limited number of displays can accept 24p deinterlaced within HD-disc machines and output at 1080/24p. Displays--except for CRT models--must also be able to deinterlace 1080/60i into 1080p60.

AIUI, most HD hardware can handle both the fractional Hz rates (so-called NTSC offset, even though HD isn't NTSC) or whole-number Hz rates. The H/DTV ATSC format has 18 frame/raster combinations, but actually 36 if you factor in the 0.1% NTSC offset. Sure I left out something, but all these parameters really takes a book. Matter of fact, a good one is C. Poynton's 2003 Digital Video and HDTV.-- John

I agree with all that, but I don't think it really answers my question. I do appreciate your input though. I'll re-read your post to make sure that the answer is not contained somewhere and I just missed it. Thanks!

Unless my math is wrong, or you repeat each frame twice for some reason, you can't get a 60p stream with that method. If you simply hold one field in a buffer and combine the next one with it, those two fields still only represent 1/60th of a second each, or 1/30th of a second for the entire frame. You would have to repeat each frame twice to get 60 fps.
AIUI, interlaced CRTs are showing standard 1080i30 at 30 fps, except the 60-fields-per-second delivery/viewing, with eyes/brain TV field merging, prevents flicker. [Prefer writing 1080/60i (30i) myself, instead of 1080i30, emphasizing the 1/60-sec TV-field 'snapshots']. Not sure where you feel the math is wrong; deinterlacing 1/60-sec fields, with a buffer-memory time delay, leads to one full frame (2 fields) displayable in 1/60-sec it seems to me. That is, dumping the buffer to the screen in 1/60 sec.

That differs, of course, from capturing 60p frames originally (ATSC's 1280X720/60p@1.5 Gbps, or the non-ATSC 1920X1080/60p@3 Gbps used for some production). -- John

AIUI, interlaced CRTs are showing standard 1080i30 at 30 fps, except the 60-fields-per-second delivery/viewing, with eyes/brain TV field merging, prevents flicker. [Prefer writing 1080/60i (30i) myself, instead of 1080i30, emphasizing the 1/60-sec TV-field 'snapshots']. Not sure where you feel the math is wrong; deinterlacing 1/60-sec fields, with a buffer-memory time delay, leads to one full frame (2 fields) displayable in 1/60-sec it seems to me. That is, dumping the buffer to the screen in 1/60 sec.

That differs, of course, from capturing 60p frames originally (ATSC's 1280X720/60p@1.5 Mbps, or the non-ATSC 1920X1080/60p@3 Gbps used for some production). -- John


I get the first part. CRT's do show 30 fps, but it is interlaced, not progressive.

What I'm asking about really pertains to fixed pixel displays (I probably should have noted that). I'm assuming "dump the buffer to the screen in 1/60 sec" means we are using a 60 Hz refresh rate, in which case we need 60 full progressive frames per second. Using the method you have described in the first paragraph (assuming we have 29.97 video material), we will run out of frames halfway through the video, and the video will be displayed at 2x. We would either have to show each frame twice, or "interpolate" half the fields so that we have 60 full frames per second instead of 30. It was my understanding that the latter is what most video mode deinterlacers do. Maybe I'm just misunderstanding what you are saying, or I just misunderstand something altogether. Please do correct me where I'm wrong, and thanks once again.

What I'm asking about really pertains to fixed pixel displays (I probably should have noted that). I'm assuming "dump the buffer to the screen in 1/60 sec" means we are using a 60 Hz refresh rate, in which case we need 60 full progressive frames per second. Using the method you have described in the first paragraph (assuming we have 29.97 video material), we will run out of frames halfway through the video, and the video will be displayed at 2x. We would either have to show each frame twice, or "interpolate" half the fields so that we have 60 full frames per second instead of 30. It was my understanding that the latter is what most video mode deinterlacers do. Maybe I'm just misunderstanding what you are saying, or I just misunderstand something altogether. Please do correct me where I'm wrong, and thanks once again.

EDIT: Material similar to this was discussed in a thread (http://archive2.avsforum.com/avs-vb/showthread.php?p=8994157&&#post8994157) with Dale Adams, who designs video processors. Recall he outlined what goes on, and I'd agree with the conclusions there if points made here differ significantly.

Maybe the different ways of looking at this relates to 1080i30, standard broadcast HD, delivered as two 1/60-sec TV fields per frame, and video capture at 30 progressive frames per sec. The latter is used for some ad production; it was also used this year for a PBS production on amateur astronomy. For broadcast and home delivery, of course, the 30 frames per second must be converted from 30 progressive frames/second to interlaced 1080/60i (30i), the standard 1080i broadcast format. This difference between 30 frames per second (fps), full progressive capture, and 30 fps delivery of 1080i seen on interlaced CRTs (seen at flicker-free 60 fields/sec) has been outlined often in various AVS threads.

60 frames per second capture/delivery/viewing arises with 1280x720/60p delivered by 720p sources like ABC, seen on progressive displays with or without scaling to another format. Most 1080i CRT displays or STBs interlace and scale such 720/60p sources to 1080/60i (30i). Some TV cameras used for interlaced 1080/60i (30i) programming capture images at 60 frames per second, but each frame must first be interlaced/converted for standard 1080/60i (30i) broadcasting.

Some newer cameras such as Sony's 1500 series capture images at 1080/60p internally but then downconvert to 720/60p, boosting the limiting resolution (http://archive2.avsforum.com/avs-vb/showthread.php?p=5667245&&#post5667245) of the 1280X720/60p format and delivering sharper images from the oversampling (1920X1080/60p). 1080/60p@3 Gbps (double the ATSC 1.5 Gbps capture rate), as mentioned earlier, is only a developing production format requiring special recording gear such as Sony's HDCAM-SR running at 800 Mbps.

Viewing 1080p60 on a fixed pixel display, deinterlaced from a standard 1080i30 image @1.5 Gps capture (74-MHz sampling), differs from viewing 1080/60p images from a HDCAM-SR, captured at 3 Gbps (148-MHz sampling) and maintained in progressive format (compressed to ~800 Mbps). -- John

Its really easy.

You get two types of material: field based and frame based.
You can encode frame based material as fields.
You do this a couple of ways ( speed it up 4% run it at 25fps and turn each frame into an odd and even field...welcome to PAL : 625/50 , or to get your 24fps rate up to 30fps you repeat fields in a 3:2 pattern to create an extra 6 frames or more accurately 12 fields if referring to 60i)

So you get :

60i material that actually consists of temporally different consecutive fields.
60i material that actually consists of 24fps material running at 23.97 fps with 3:2 pulldown patterned repeat fields to take us up to 60i.

So how do you ideally deinterlace it?

the latter case (3:2 pulldowned 60i) is actually quite easy as assuming there are no bad edits the deinterlacer just has to watch for repeat fields at the right time : strip them out : then repeat the resulting 24fps sequence at a rate to meet the desired display: ie 72Hz each frame gets repeated 3 times: if a non-multiple of 24 refresh rate is required then some frames get repeated more than others albeit regular full frame repeat patterns but you're essentally back to judder.

Like I said 3:2 pulldown is comparatively easy to detect and remove unless there are bad edits which requires the deinterlacer to have to reset and identify the 3:2 pulldown pattern again ( the cadence).

The good thing about this sort of deinterlace scenario is that you are essentially reconstructing real frames with no interpolation.

It gets messy when you have 60i (or 50i) native field based material to display on a progressive display device. The easiest way is to take each field and resize it ( interpolate) to be a full size frame and then display at 60fps. The good things about this type of deinterlace is it doesn't have any interlace artifacts even if there are bad edits in the material and you don't lose any information held in the fields themselves: however fine horizontal detail will appear to "bob" from frame to frame ( as it only appears on every other field originally). Hence its known as a "bob" deinterlace ( or scalar bob).

Then you get motion adaptive deinterlacing which is often overmystified but simply means the deinterlacer will identify when motion occurs in the image and switch between a simple "weave" deinterlace ( takes two fields reweaves them into a frame) and a bob based on detected movement. When the image is static you get 2 fields worth of information in each frame so it looks as sharp as possible but when the image moves the deinterlace drops to a bob. This is good because you get maximum sharpness on static imagery and no interlace artifacts on moving imagery and hopefully the softening from the bob will be less noticable on the moving sequences. If the deinterlacer doesn't switch quickly enough you will see mismatched field artifacts . Mixed cadence material ( news ticker tapes infamously) will trip this sort of deinterlacer up.

Then you get motion compensated deinterlacers which use "frame segmentation" techniques : think of this as the bob/weave switch happening across different areas of the same frame rather than the full frame. This is quite a sophisticated deinterlace and has only been available for a few years in the consumer marketplace ( ATi and Nvidia use this technique on some cards with mixed success/support).However often its not actually that much better looking than a motion adaptive deinterlacer.

50i (PAL) cadence detection is actually quite tricky compared with 60i as the 3:2 pulldown pattern sticks out like a sore thumb in signal terms whereas its quite tricky to pickup on a PAL 2:2 pulldown pattern consistently.

And to be honest thats about it as far as deinterlacing is concerned.

My concern is with HD DVD players that only output 1080i... could the target display mistake the 1080i59.97 as video instead of film and apply a video mode deinterlace? It seems as though this would produce a smeared gray image or something that flickers. The same question would apply to players that only output 1080p29.97, can we count on the player to deinterlace correctly?

I can give you a real world example of what happens if a player or TV does not deinterlace. The PS3 does not deinterlace and bobs the 1080i stream if you force 1080p output (uncheck 1080i in the video options). You get a horrific flashing screen using the vertical resolution pattern on your calibration disc. Unfortunately bobbing is by far the most common behavior you will encounter when a progressive based display is fed a 1080i signal. For the more recent HDTVs that do weave a 1080i signal back to 1080p, many still will only do 30p and do not do 3:2 pull down (like my Sony 70" XBR2). There are a growing number displays that will detect and weave back to 24p, but it's not common enough in the HDTV world that you can count upon your average display doing it yet.

First, I apologize for wording my question so poorly.. I think that has led to a lot of confusion about what I was asking.

The pattern I am describing on AVS HD is currently 29.97 fps interlaced video (not film of course). I was having trouble understanding how some deinterlacers (such as the XA2) were apparently resolving the pattern correctly, while others (apparently the PS3) would flash or blink. I was thinking that all video mode deinterlacers should have those flash or blink (possibly just a gray screen also) type problems with this pattern, mainly because I thought all of them would attempt to interpolate one field for each frame (yielding 60 fps progressive). Mr. D explained that type of deinterlacing as a "bob" deinterlacer.

His next paragraph, quoted below, likely explains how the XA2 is able to deinterlace the pattern properly. I never really understood what a motion adaptive deinterlacer did, but now that I do, I suspect the XA2 uses one and, as a result, displays this type of pattern cleanly.


Then you get motion adaptive deinterlacing which is often overmystified but simply means the deinterlacer will identify when motion occurs in the image and switch between a simple "weave" deinterlace ( takes two fields reweaves them into a frame) and a bob based on detected movement. When the image is static you get 2 fields worth of information in each frame so it looks as sharp as possible but when the image moves the deinterlace drops to a bob. This is good because you get maximum sharpness on static imagery and no interlace artifacts on moving imagery and hopefully the softening from the bob will be less noticable on the moving sequences. If the deinterlacer doesn't switch quickly enough you will see mismatched field artifacts . Mixed cadence material ( news ticker tapes infamously) will trip this sort of deinterlacer up.

The XA2 would realize the image is static, and simply put "weave" two coded fields together to get a frame, yielding a completely clean pattern. The PS3 would still attempt to interpolate, which causes havoc with this type of pattern.

Now to go a bit further.

1) Assuming we want this pattern to be displayed cleanly on as many playback/display combinations as possible, would it be better to encode it as 24 fps (interlaced) with pulldown? I guess the real question is whether or not we should assume that most players will do a film mode deinterlace better than a video mode deinterlace. It is preferable to have the pattern interlaced because it makes it much easier for the MPEG2 encoder to encode, and reduces the file size considerably. So the options are 29.97 interlaced "video," or basically faking 23.967 fps "film" with pulldown.

2) Question for Mr.D (and anyone else)... if a static image is encountered by an adaptive deinterlacer, how does it get to 60 fps? It just shows each frame twice?

Thanks a lot for all the input.

I can give you a real world example of what happens if a player or TV does not deinterlace. The PS3 does not deinterlace and bobs the 1080i stream if you force 1080p output (uncheck 1080i in the video options). You get a horrific flashing screen using the vertical resolution pattern on your calibration disc. Unfortunately bobbing is by far the most common behavior you will encounter when a progressive based display is fed a 1080i signal. For the more recent HDTVs that do weave a 1080i signal back to 1080p, many still will only do 30p and do not do 3:2 pull down (like my Sony 70" XBR2). There are a growing number displays that will detect and weave back to 24p, but it's not common enough in the HDTV world that you can count upon your average display doing it yet.
The flashing pattern with a test DVD reads like what Gary Merson discovered, widely posted, with many HD displays. Three of his articles, sublinked, are here (http://archive2.avsforum.com/avs-vb/showthread.php?p=9178202#post9178202) and include TV model lists.

So-called 540p bobbing, the technique used, simply doubles each 1/60-sec, 540-line TV field or half-frame to 1080p60 frames. This can halve vertical resolution since the frames are only derived from half-fames. Believe Merson used a HQ Video beta-test disc not available while researching his articles, but it and other sources of the alternating B&W-line SMPTE-type patterns are also available as HD discs. AIUI, the flashing pattern represents black, then white, lines being alternately doubled to full frames instead of being displayed properly as stacked B&W lines.

"2) Question for Mr.D (and anyone else)... if a static image is encountered by an adaptive deinterlacer, how does it get to 60 fps? It just shows each frame twice?
The 540p bobbing technique, outlined and exampled above, gets to 60 fps by just doubling 1/60-sec, 540-line TV fields. Video-processor engineer Dale Adams, in the thread I linked above, indicated that 1/60-sec frames are not obtained by doubling 1080i30 'frames'. Static-image 1080i30 lines can just use a simple 'weave' technique since there's no movement in the time-seperated TV fields. I can picture deinterlacing TV fields involving movement in a buffer memory and dumping complete 1/60-sec frames to a fixed pixel display each 1/60 sec. But guess I'll leave it to others to outline how 1/60-sec frames results from non-static 1/60-sec fields, (not meaning the whole adaptive deinterlacing technique) --or offer alternative explanations. -- John

The flashing pattern with a test DVD reads like what Gary Merson discovered, widely posted, with many HD displays. Three of his articles, sublinked, are here (http://archive2.avsforum.com/avs-vb/showthread.php?p=9178202#post9178202) and include TV model lists.

So-called 540p bobbing, the technique used, simply doubles each 1/60-sec, 540-line TV field or half-frame to 1080p60 frames. This can halve vertical resolution since the frames are only derived from half-fames. Believe Merson used a HQ Video beta-test disc not available while researching his articles, but it and other sources of the alternating B&W-line SMPTE-type patterns are also available as HD discs. AIUI, the flashing pattern represents black, then white, lines being alternately doubled to full frames instead of being displayed properly as stacked B&W lines.


The 540p bobbing technique, outlined and exampled above, gets to 60 fps by just doubling 1/60-sec, 540-line TV fields. Video-processor engineer Dale Adams, in the thread I linked above, indicated that 1/60-sec frames are not obtained by doubling 1080i30 'frames'. Static-image 1080i30 lines can just use a simple 'weave' technique since there's no movement in the time-seperated TV fields. I can picture deinterlacing TV fields involving movement in a buffer memory and dumping complete 1/60-sec frames to a fixed pixel display each 1/60 sec. But guess I'll leave it to others to outline how 1/60-sec frames results from non-static 1/60-sec fields, (not meaning the whole adaptive deinterlacing technique) --or offer alternative explanations. -- John

Thanks for all the links John, I have some interesting reading to do...

Now to go a bit further.

1) Assuming we want this pattern to be displayed cleanly on as many playback/display combinations as possible, would it be better to encode it as 24 fps (interlaced) with pulldown?.

That really depends what you want the pattern to tell you about the deinterlacer: cadence detection , interpolation quality , bad edit recovery...




2) Question for Mr.D (and anyone else)... if a static image is encountered by an adaptive deinterlacer, how does it get to 60 fps? It just shows each frame twice?



Yep. The deinterlacer in this mode isn't concerned with 3:2 pulldown as its keyed on motion. Ideally a good deinterlacer will check for 3:2 and 2:2pulldown cadence before dropping into the field mode anyway before resorting to a motion adaptive mode .

When its weaving on static imagery it will double the frame rate. When its bobbing it will match the field rate. They normally look at a few fields to detect the cadence ( it sounds quick but the time intervals involved for today's hardware is more than long enough for quite sophisticated processing.

Just think of the deinterlacing types themselves as quite simple : the smartish bit is the motion adaptation and 3:2 pulldown detection (although that's quite easy).

That really depends what you want the pattern to tell you about the deinterlacer: cadence detection , interpolation quality , bad edit recovery...





Yep. The deinterlacer in this mode isn't concerned with 3:2 pulldown as its keyed on motion. Ideally a good deinterlacer will check for 3:2 and 2:2pulldown cadence before dropping into the field mode anyway before resorting to a motion adaptive mode .

When its weaving on static imagery it will double the frame rate. When its bobbing it will match the field rate. They normally look at a few fields to detect the cadence ( it sounds quick but the time intervals involved for today's hardware is more than long enough for quite sophisticated processing.

Just think of the deinterlacing types themselves as quite simple : the smartish bit is the motion adaptation and 3:2 pulldown detection (although that's quite easy).

Thanks Mr. D:)

Originally Posted by hwjohn
2) Question for Mr.D (and anyone else)... if a static image is encountered by an adaptive deinterlacer, how does it get to 60 fps? It just shows each frame twice?
The '06 post (http://archive2.avsforum.com/avs-vb/showthread.php?p=8962880&&#post8962880) also covering this ground, which I cited earlier, seems to differ with a 'showing-each-frame-twice' conclusion, with video-processor-designer Dale Adams disagreeing with this concept. -- John.

The '06 post (http://archive2.avsforum.com/avs-vb/showthread.php?p=8962880&&#post8962880) also covering this ground, which I cited earlier, seems to differ with a 'showing-each-frame-twice' conclusion, with video-processor-designer Dale Adams disagreeing with this concept. -- John.


If the deinterlacer decides to weave two adjacent fields together its essentially generating 30fps from 60i. If this needs to be displayed at greater frame rate then it will simply repeat frames as required : either all of them twice to reach 60Hz or some more than others to reach a non 30 multiple.
It can't do anything else in this mode. If it then switches to a bob it will be generating 60fps anyway so no repeats in this particular mode with a 60Hz refresh.

The deinterlacer is either going to generate 24fps , 30 fps or 60 fps and then repeat frames as required to reach the display rate. There are also scenarios where an unwise refresh rate will force the culling of frames : forcing a 24fps refresh multiple with 30fps material, this gives rise to massive judder.

The frame repeat/culling is a function of the end display rate , the actual discrete deinterlacer stage isn't necessarily concerned with this.

So you get :

60i material that actually consists of temporally different consecutive fields.
60i material that actually consists of 24fps material running at 23.97 fps with 3:2 pulldown patterned repeat fields to take us up to 60i.

So how do you ideally deinterlace it?

the latter case (3:2 pulldowned 60i) is actually quite easy as assuming there are no bad edits the deinterlacer just has to watch for repeat fields at the right time : strip them out : then repeat the resulting 24fps sequence at a rate to meet the desired display: ie 72Hz each frame gets repeated 3 times: if a non-multiple of 24 refresh rate is required then some frames get repeated more than others albeit regular full frame repeat patterns but you're essentally back to judder.


Yes - it is common for 24fps material to be encoded in 1080/60i using 3:2 field pull down. This is then often required to be displayed on a progressive 60Hz display. To do this the 3:2 field-pulldown is detected and removed, delivering a 1080/24p signal that is then frame repeated at 3:2 (this time for full 1080p frames, not 1080i 540 line fields) Result still has 3:2 cadence and 3:2 judder, but has full progressive resolution and is fully de-interlaced.

As to your other comments elsewhere in the thread about 2:2 vs 3:2 detection - absolutely.

Not sure if it has been mentioned - but modern high-end de-interlacing can do a lot more than a bob/weave switch on movement at the field/frame level. Some do it on a block-by-block or pixel-by-pixel level, and very high-end models try to motion-track - retaining high resolution detail on movement. Some of the best de-interlacers available use phase correlation techniques I believe (as used on high-end standards converters?)

If the deinterlacer decides to weave two adjacent fields together its essentially generating 30fps from 60i. If this needs to be displayed at greater frame rate then it will simply repeat frames as required : either all of them twice to reach 60Hz or some more than others to reach a non 30 multiple.
It can't do anything else in this mode. If it then switches to a bob it will be generating 60fps anyway so no repeats in this particular mode with a 60Hz refresh.



Though in implementation terms - a de-interlacer that switches dynamically between bob and weave based on motion (particularly dynamically within a frame rather than uniformly for the entire frame) working at 60p output will presumably be "output based" - and thus when weaving would implement the repetition as part of the de-interlacing stage, rather than after it, so that it runs at the same rate as bobbed material ?


The deinterlacer is either going to generate 24fps , 30 fps or 60 fps and then repeat frames as required to reach the display rate.


But this rate is likely to be fixed at a single frame rate - even when a mix of Bob and Weave is used dynamically. Meaning that some repetition will be part of the de-interlacer not afterwards.


There are also scenarios where an unwise refresh rate will force the culling of frames : forcing a 24fps refresh multiple with 30fps material, this gives rise to massive judder.


Yes - where the output is fixed at 1080/24p you can't watch 1080/60i native material without (in my view) intolerable judder...


The frame repeat/culling is a function of the end display rate , the actual discrete deinterlacer stage isn't necessarily concerned with this.

Though in 60p (or 50p) de-interlacing of 1080/60i (or 50i) native material the frame repetition for WEAVE de-interlaced content may be, so that it retains the same frame rate as BOB stuff in the same stream. (Particularly if the Bob/Weave decision is taken at the block or pixel level rather than the frame level?)

Though in 60p (or 50p) de-interlacing of 1080/60i (or 50i) native material the frame repetition for WEAVE de-interlaced content may be, so that it retains the same frame rate as BOB stuff in the same stream. (Particularly if the Bob/Weave decision is taken at the block or pixel level rather than the frame level?)

I do mention frame segmentation techniques in a previous post but was sticking to how deinterlacing relates to frame rate. Motion compensated deinterlacers are hopefully something that kicks in only once the deinterlacer has failed to detect a 3:2 or 2:2 cadence in exactly the same way that a motion adaptive would.

Regardless of deinterlace type I'm sure the frame repeat is merely a function of the frame buffer rather than having any necessary carry through to the deinterlace stage itself. The deinterlacer merely creates a progressive sequence , the framebuffer then displays it as required by the output refresh rate.

Not sure if it has been mentioned - but modern high-end de-interlacing can do a lot more than a bob/weave switch on movement at the field/frame level. Some do it on a block-by-block or pixel-by-pixel level, and very high-end models try to motion-track - retaining high resolution detail on movement. Some of the best de-interlacers available use phase correlation techniques I believe (as used on high-end standards converters?)

Indeed, anything doing 'bob' or 'weave' decoding is about 20 years out of date! :) There's a good round up of the methods here: http://downloads.bbc.co.uk/rd/pubs/reports/1996-09.pdf , not all of them are intuitive for people who tend to come from the computing side... Especially this one, http://neuron2.net/misc/USP4789893.pdf which is at the heart of Snell and Wilcox's ARC150.

Steven

Motion compensated deinterlacers are hopefully something that kicks in only once the deinterlacer has failed to detect a 3:2 or 2:2 cadence in exactly the same way that a motion adaptive would.

I'd hope they don't -- it is very common (at least in the UK) for a frame to contain both progressive material with a 2:2 cadence and interlaced material. I've seen deinterlacers that do what you suggest go haywire with that sort of material. Indeed, I'm pretty certain there is one episode of 'Spooks' (MI5 in the US) that has on screen at the same time interlaced material, and two sets of progressive footage who's cadence is out of phase with each other... As the footage was DVE'd about the screen the deinterlacer kept locking from source to source.

I've seen similar stuff with 2:3 pulldown material too editted in the 60i domain (Buffy springs to mind).

Steven

I'd hope they don't -- it is very common (at least in the UK) for a frame to contain both progressive material with a 2:2 cadence and interlaced material. I've seen deinterlacers that do what you suggest go haywire with that sort of material. Indeed, I'm pretty certain there is one episode of 'Spooks' (MI5 in the US) that has on screen at the same time interlaced material, and two sets of progressive footage who's cadence is out of phase with each other... As the footage was DVE'd about the screen the deinterlacer kept locking from source to source.

I've seen similar stuff with 2:3 pulldown material too editted in the 60i domain (Buffy springs to mind).

Steven

An MPEG2 frame can have progressive and interlaced "parts" in the same frame?

An MPEG2 frame can have progressive and interlaced "parts" in the same frame?

You can't rely on the flags in the MPEG2 stream -- they are often wrong!

Steven

I'd hope they don't -- it is very common (at least in the UK) for a frame to contain both progressive material with a 2:2 cadence and interlaced material. I've seen deinterlacers that do what you suggest go haywire with that sort of material


This is hardly the fault of the deinterlacer. You will always get situations that can break any deinterlacer. I do mention the classic case of News ticker tapes with frame based cadence breaking adaptive deinterlacers.

I'd much rather have a deinterlacer that stuck with a suitable weave mode when it recognised frame originated material albeit with the risk of failing on badly put together material than a deinterlacer that stuck with motion compensated segmented frame techniques for everything.

Indeed, anything doing 'bob' or 'weave' decoding is about 20 years out of date! :) There's a good round up of the methods here: http://downloads.bbc.co.uk/rd/pubs/reports/1996-09.pdf , not all of them are intuitive for people who tend to come from the computing side... Especially this one, http://neuron2.net/misc/USP4789893.pdf which is at the heart of Snell and Wilcox's ARC150.

Steven

And what exactly is wrong with a weave deinterlace for frame based material?

You can't rely on the flags in the MPEG2 stream -- they are often wrong!

Steven

I realize that. But you seemed to suggest that you can have an MPEG2 stream with interlaced and progressive "parts" in a single frame (not stream)?

And what exactly is wrong with a weave deinterlace for frame based material?

Nothing as such -- if you can guarantee that the cadence will not change and you'll never get anything but 2:2 or 2:3, which you can't guarantee -- (remember video fields are discretely entities and how they are stored varies from device to device even if they use the same codec -- if I feed the same 25psf video source to my Panasonic and Sony DV VCRs they package up the fields with a different cadence...) It's less of a problem with movie transfers I'll admit (but then you should be storing them as 24p on the disk and adding 3:2/2:2 pulldown in the player but that is another debate), but is quite likely with other types of material. TV drama for instance -- which in the UK is likely to be edited as 25i, not 25psf, for valid reasons. It'd only take a Varispeeded clip to change the phase of a shot's cadence. Or 25i overlayed graphics that don't trigger the deinterlacer's threshold level to switch to a different mode.

And that's why weave is a naive solution, it requires a) being able to detect progressive segmented frame material, b) detect the cadence of the segmented frames, and when it changes and c) weave the fields together in the correct order. (c) is the easy part, (a) and (b) are hard -- and so it is better to find a deinterlacing algorithm that is functionally equivalent to weave but doesn't require (a) and (b) as such discrete components.


I realize that. But you seemed to suggest that you can have an MPEG2 stream with interlaced and progressive "parts" in a single frame (not stream)?

Just because the MPEG2 frames are tagged as progressive and/or interlace doesn't mean the video content inside them is progressive or interlaced :) I have DVDs of very video encoded content that are encoded as progressive frames...

Steven

Just because the MPEG2 frames are tagged as progressive and/or interlace doesn't mean the video content inside them is progressive or interlaced :) I have DVDs of very video encoded content that are encoded as progressive frames...

Steven

Sure, but you can't have a single frame that is partially interlaced or partially progressive in MPEG2.

Sure, but you can't have a single frame that is partially interlaced or partially progressive in MPEG2.

But the material in it can be, that's the point. If you have an image that contains 66% progressive material and 33% interlaced you could well be better off compressing that frame as if it were progressive than as interlaced -- the only difference is the way the DCT blocks are constructed (either as a single 8x8 block or as 2 8x4 blocks). The encoder still takes two consecutive fields and interleaves the lines together and the decoder does the reverse.

Hence the material in the MPEG2 frame doesn't have to correspond to what the flag says (it's just inefficient) and I know of commercial DVD releases of interlaced material that have been encoded as progressive frames.

Steven

But the material in it can be, that's the point.

To me this thread seems rather off-topic compared to the original intent of where hwjohn must have been going. For our purpose (http://www.avsforum.com/avs-vb/showthread.php?t=948496), a more practical discussion might be:

1) How can we create a 24p mpeg2 file that is HD DVD compliant? (requires Avisynth compatible encoder)
2) How can we encode or transcode to a 24p H264 m2ts file with ac3 audio that is AVCHD compliant using software that doesn't exceed a few hundred dollars? (H264 encode/transcode, xmuxer, and tsremux seemed the best lead but early test files had video error messages when importing into DVD MovieFactory 6 Plus)(Sony Vegas seemed like another lead, but the trial doesn't include m2ts transcoding to test)

To me this thread seems rather off-topic compared to the original intent of where hwjohn must have been going. For our purpose (http://www.avsforum.com/avs-vb/showthread.php?t=948496), a more practical discussion might be:

1) How can we create a 24p mpeg2 file that is HD DVD compliant? (requires Avisynth compatible encoder)
2) How can we encode or transcode to a 24p H264 m2ts file with ac3 audio that is AVCHD compliant using software that doesn't exceed a few hundred dollars? (H264 encode/transcode, xmuxer, and tsremux seemed the best lead but early test files had video error messages when importing into DVD MovieFactory 6 Plus)(Sony Vegas seemed like another lead, but the trial doesn't include m2ts transcoding to test)

We can do #1 (at least the tests have worked so far). #2... not so much ;)

It has gotten a bit off topic, but I think Mr. D already answered my original question for the most part (mainly why your XA2 deinterlaced the horizontal line pattern when other players couldn't). Right now I guess we are just discussing deinterlacing techniques for the heck of it.

For what it's worth, my BDP-S1 also deinterlaces the same pattern correctly. My computer doesn't deinterlace correctly. I haven't tried to see how the TV would deinterlace with either standalone player.