hello everybody… :wink:
I'm working with some HD content…specifically:
1920 x 1080 (often called 1080i)
When I am comping it, the pixel aspect ratio is Non-square (1.33:1)
So my question is…should I change the pxel aspect ratio in houdini? Or just use square pixels and then stretch them when comping?
My other quesition is, if I am going to change the ratio beforehand, would I just change it to 1.33?
thanks!
Full Version: pixel aspect ratios
I'm not sure which is the better way but just want to point out that 1080i is a 16:9 (1.77) aspect, not 4:3 (1.33). 
ah….thats probably where my problem is….you see my compositor probably is converting it to 1.33….which is where I got that from.
thanks!
thanks!
just to point out not sure if I understand problem correctly…
for D1 we have 720/576*1.0667 = 1.333 = 4/3,
and for 1080i - where pixel is square: 1920/1080*1 = 1.777 means = 16/9.
And you should definitely always render with proper pixel aspect ratio unless you want your edges look like rendered without antialissing
cheers,
SY.
not sure if I understand problem correctly… for D1 we have 720/576*1.0667 = 1.333 = 4/3,
and for 1080i - where pixel is square: 1920/1080*1 = 1.777 means = 16/9.
And you should definitely always render with proper pixel aspect ratio unless you want your edges look like rendered without antialissing
cheers,
SY.
English is a great language, but there's some wonderfully confusing holes. 
There's a significant difference between *pixel* aspect ratio and “aspect ratio”, and they're commonly mixed up in conversation, as in this thread. The former is related to the physical characteristics of the target display device, the latter is the ratio of width to height of the image.
As related here [adelphia.com] in engineer-free language, or here [en.wikipedia.org] in rascally link-heavy wikipedia-speak, 1080i has *square* pixels, not rectangular. Essentially, the pixel aspect is unrelated to the image aspect ratio(of course it's *related*, but they must be considered divorced of one another). It's got to do with those physical pixels that are on your TV set - *not* the overall terms like “16:9”. When you look at a serial digital video image of a perfect circle on your computer monitor, it should looked stretched horizontally as an ellipse(unless you compensate in your viewer). It *must* leave the compositing chain to video looking like that. 1080i must be square, in other words a circle on your monitor will be a circle on the final display, not an ellipse - just like film(which has no pixels).
I still have issues with the way the Houdini compositor embeds pixel aspect ratio(as Mark well knows ) - it's *only* related to the physical characteristics of the end display unit, and shouldn't be inherently part of the compositing process. A 720x486 0.9 PAR image is digitally *precisely* the same whether or not you screen it on your monitor or on a video screen - the display device determines how it physically looks to the end viewer. There's nothing wrong with saying your project is in 0.9 PAR, and the viewer compensates for that(which I *think* it does) as a display process only for the convenience of the compositor, but mixing together images in a composting network of different aspect ratios has no real meaning, except that you've got some sort of behind the scenes filtering going on that's out of your control depending on the image. Shouldn't be - if you need to pull in an SD sequence to somehow embed within a 1080i image, you should have upfront control *per image* of how that's filtered when resized. You can do that in Houdini, but you need to switch off all this PAR per sequence stuff and work in square. Don't be misguided by it!
Anyway, pet peeve. However, much of the discussion here has been based on a misunderstanding of the process, thought I should set it straight. If I've been unclear, ask away…
Cheers,
J.C.
There's a significant difference between *pixel* aspect ratio and “aspect ratio”, and they're commonly mixed up in conversation, as in this thread. The former is related to the physical characteristics of the target display device, the latter is the ratio of width to height of the image.As related here [adelphia.com] in engineer-free language, or here [en.wikipedia.org] in rascally link-heavy wikipedia-speak, 1080i has *square* pixels, not rectangular. Essentially, the pixel aspect is unrelated to the image aspect ratio(of course it's *related*, but they must be considered divorced of one another). It's got to do with those physical pixels that are on your TV set - *not* the overall terms like “16:9”. When you look at a serial digital video image of a perfect circle on your computer monitor, it should looked stretched horizontally as an ellipse(unless you compensate in your viewer). It *must* leave the compositing chain to video looking like that. 1080i must be square, in other words a circle on your monitor will be a circle on the final display, not an ellipse - just like film(which has no pixels).
I still have issues with the way the Houdini compositor embeds pixel aspect ratio(as Mark well knows
) - it's *only* related to the physical characteristics of the end display unit, and shouldn't be inherently part of the compositing process. A 720x486 0.9 PAR image is digitally *precisely* the same whether or not you screen it on your monitor or on a video screen - the display device determines how it physically looks to the end viewer. There's nothing wrong with saying your project is in 0.9 PAR, and the viewer compensates for that(which I *think* it does) as a display process only for the convenience of the compositor, but mixing together images in a composting network of different aspect ratios has no real meaning, except that you've got some sort of behind the scenes filtering going on that's out of your control depending on the image. Shouldn't be - if you need to pull in an SD sequence to somehow embed within a 1080i image, you should have upfront control *per image* of how that's filtered when resized. You can do that in Houdini, but you need to switch off all this PAR per sequence stuff and work in square. Don't be misguided by it!Anyway, pet peeve.
However, much of the discussion here has been based on a misunderstanding of the process, thought I should set it straight. If I've been unclear, ask away…Cheers,
J.C.
JColdrick… hmm…. you are the master, I'm little poster here but somehow I have to not agree with you or it's just my English misleads my thoughts what is as you said very propable 
ops:
Of course pixel aspect ratio is a different part of story then image aspect but it doesn't mean that one should leave it in rendering process and render in square pixels… is that what you said? For any format with PAR 1:1 this is true but for others not. It's hard to explain in words what is going on - especially that I'm not even sure if we have some controvery here.
But to show it visually:
If one got footage in 2K (4:3 - 2048x1556) from telecine (which has square pixeles - of course) and do compositing for TV in PAL:
this shouldn't happen:
2048 * 1/ (2048/720) ~ 720 (I'm dividing this way intentionally)
1556 * 1/ (1556/576) ~ 576 -> PAL
because as you know the image will be stretched <–> on TV divice.
this should happen:
2048 * 1/ (2048/720) ~ 720 — as above
but:
1556 * 1/(1556/576) * PAR…. so image is resized to fit X but not Y. This way you image is stretched >–< on computer screen but on TV device a circle is a circle.
This way video cameras work, it “see” some more world on left and right side of view and put it inside a frame. Exacly like anamorphic lenses.
I think it's better to let matra filter pixels in respet to this then do stretching in compo… am I wrong? or do I understand you in a wrong way?
Hope I'm not in some terrible mistake here… :roll:
cheers,
SY.
ops: Of course pixel aspect ratio is a different part of story then image aspect but it doesn't mean that one should leave it in rendering process and render in square pixels… is that what you said? For any format with PAR 1:1 this is true but for others not. It's hard to explain in words what is going on - especially that I'm not even sure if we have some controvery here.
But to show it visually:
If one got footage in 2K (4:3 - 2048x1556) from telecine (which has square pixeles - of course) and do compositing for TV in PAL:
this shouldn't happen:
2048 * 1/ (2048/720) ~ 720 (I'm dividing this way intentionally)
1556 * 1/ (1556/576) ~ 576 -> PAL
because as you know the image will be stretched <–> on TV divice.
this should happen:
2048 * 1/ (2048/720) ~ 720 — as above
but:
1556 * 1/(1556/576) * PAR…. so image is resized to fit X but not Y. This way you image is stretched >–< on computer screen but on TV device a circle is a circle.
This way video cameras work, it “see” some more world on left and right side of view and put it inside a frame. Exacly like anamorphic lenses.
I think it's better to let matra filter pixels in respet to this then do stretching in compo… am I wrong? or do I understand you in a wrong way?
Hope I'm not in some terrible mistake here… :roll:
cheers,
SY.
JColdrick
There's a significant difference between *pixel* aspect ratio and “aspect ratio”, and they're commonly mixed up in conversation, as in this thread.
Sorry for the confusion. I totally missed the part about pixel aspect ratio. I took a quick look and didn't see an obvious answer. What are the common situations where we run into non-square pixel aspect ratios then? Off the top of my head, I can see us getting into that whenever our video card's aspect ratio is different from our monitor's.
Lol, I'm hardly a master. Here's the basic issue - pixels are those little coloured dots you look at - and there are some devices - video display units - that don't have perfectly square pixels. If you look closeup at a good video monitor you can see that. However, pretty well all other devices, like your computer monitor, HD screens, etc all have square pixels. Since we have to deal in the smallest visual unit when outputting to a device, the end image has to be in pixels - however what you're seeing on your monitor won't match what you will see in video because that *display* unit draws it differently. This is why a serial digital NTSC image is 720x486 @ 0.9 PAR - it compensates for that difference so a perfect circle rendered at that res will look “wrong” on your computer screen(because of the camera PAR setting), but will look right when sent to some sort of video buffer. If it was HD, it must be rendered the same as your computer screen - @ 1 PAR - because that's the HD standard. It's all tied to the messy business of how video display units were created and altered over time.
As far as the plates you get - again it's all about the *display devices* - not the camera or lenses. I'm still just talking about pixel aspect ratio. However, I'm unclear about the plates example you use. Are you using a film plate that will go to PAL? In that case, all compositing *and* rendering should be at a PAR of 1, then if you want to go to PAL, you would resize to a PAR of 0.9/1.1 in the final output, yes. But I assume the reason you have hires plates is because you want to work in hires - which is a PAR of 1, if it came from film.
Again, I'm not certain of your pipeline, so perhaps that's not the way it's working. The simplest solution would be to shoot a circle, run it through the pipeline, and see what you get.
Cheers,
J.C.
Here's the basic issue - pixels are those little coloured dots you look at - and there are some devices - video display units - that don't have perfectly square pixels. If you look closeup at a good video monitor you can see that. However, pretty well all other devices, like your computer monitor, HD screens, etc all have square pixels. Since we have to deal in the smallest visual unit when outputting to a device, the end image has to be in pixels - however what you're seeing on your monitor won't match what you will see in video because that *display* unit draws it differently. This is why a serial digital NTSC image is 720x486 @ 0.9 PAR - it compensates for that difference so a perfect circle rendered at that res will look “wrong” on your computer screen(because of the camera PAR setting), but will look right when sent to some sort of video buffer. If it was HD, it must be rendered the same as your computer screen - @ 1 PAR - because that's the HD standard. It's all tied to the messy business of how video display units were created and altered over time.As far as the plates you get - again it's all about the *display devices* - not the camera or lenses. I'm still just talking about pixel aspect ratio. However, I'm unclear about the plates example you use. Are you using a film plate that will go to PAL? In that case, all compositing *and* rendering should be at a PAR of 1, then if you want to go to PAL, you would resize to a PAR of 0.9/1.1 in the final output, yes. But I assume the reason you have hires plates is because you want to work in hires - which is a PAR of 1, if it came from film.
Again, I'm not certain of your pipeline, so perhaps that's not the way it's working. The simplest solution would be to shoot a circle, run it through the pipeline, and see what you get.
Cheers,
J.C.
I think we understand each other c well… and I catch dots/pixel reality quite well also . But still there are two points:
(1) mantra IS like virutal camera and should take picture in target format. So if circle shot with phisical camera appears on computer monitor stretched - so circle from mantra (if renderd for TV) should be stretched and it's definitely better NOT to do it in post.
(2) as I know pixel ratio has nothing to do with property of phisical devices, but more with former media digitalization. We used to have 768x576 for PAL with square pixels - or to be precise - with no pixels but lines of analog signal. Since these number are difficult for managing in digital postprocess/broadcast for some reason, they were changed on 720x576 and difference between screen aspects are compensated with pixel format. This means much more then actual shape of device dots - this means that the world IS recorded in non-square pixels (on video) and once recorded it's not so easy to change it without loss of quality, jittering edges etc.
My example treating 2K plate prevents a circle shot on stock camera to apear stretched in TV. Images from mantra rendered with square pixels will be stretched on TV although they can keep their shape if rendered with proper pixel ratio. Do we agree?
. But still there are two points:(1) mantra IS like virutal camera and should take picture in target format. So if circle shot with phisical camera appears on computer monitor stretched - so circle from mantra (if renderd for TV) should be stretched and it's definitely better NOT to do it in post.
(2) as I know pixel ratio has nothing to do with property of phisical devices, but more with former media digitalization. We used to have 768x576 for PAL with square pixels - or to be precise - with no pixels but lines of analog signal. Since these number are difficult for managing in digital postprocess/broadcast for some reason, they were changed on 720x576 and difference between screen aspects are compensated with pixel format. This means much more then actual shape of device dots - this means that the world IS recorded in non-square pixels (on video) and once recorded it's not so easy to change it without loss of quality, jittering edges etc.
My example treating 2K plate prevents a circle shot on stock camera to apear stretched in TV. Images from mantra rendered with square pixels will be stretched on TV although they can keep their shape if rendered with proper pixel ratio. Do we agree?
I agree with #1. Do it all the time. The original post was asking about HD, not SD video, though!
#2, ummm, yes, but frankly I almost never work with DV video cam work. Mostly everything we work with is shot on film and transferred to SD video or transferred at HD or film resolution. If you're shooting *HD* cam work, you work at PAR of 1 - that's the standard(see my previous links). So I'm still not sure I agree about the specific example you're talking about. You're getting a 2K image, right? That means it was shot HD or film, right? The PAR for HD is 1, not 0.9.
I think we basically agree, it's just the details that aren't communicating properly. I do know that PAR is definitely related to the display device, though. I understand in the end *everything* is analog, but it's got to do with the decisions made by the engineers when the need came about to digitize the analog signal - they had to make some decisions, and we're stuck with them. I'm not sure I follow the bit about “difference between screen aspects”. Don't forget - a pixel is a pixel - it's just a set of values decribing colour of a point on a screen, nothing more. It doesn't itself have an “aspect”. Only the display device has a pixel aspect ratio, and we shoot/render to accomodate for that.
Cheers,
J.C.
#2, ummm, yes, but frankly I almost never work with DV video cam work. Mostly everything we work with is shot on film and transferred to SD video or transferred at HD or film resolution. If you're shooting *HD* cam work, you work at PAR of 1 - that's the standard(see my previous links). So I'm still not sure I agree about the specific example you're talking about. You're getting a 2K image, right? That means it was shot HD or film, right? The PAR for HD is 1, not 0.9.
I think we basically agree, it's just the details that aren't communicating properly. I do know that PAR is definitely related to the display device, though. I understand in the end *everything* is analog, but it's got to do with the decisions made by the engineers when the need came about to digitize the analog signal - they had to make some decisions, and we're stuck with them. I'm not sure I follow the bit about “difference between screen aspects”. Don't forget - a pixel is a pixel - it's just a set of values decribing colour of a point on a screen, nothing more. It doesn't itself have an “aspect”. Only the display device has a pixel aspect ratio, and we shoot/render to accomodate for that.
Cheers,
J.C.
“… I'm not sure I follow the bit about ”difference between screen aspects“….”
that's the point. I wrote “screen aspects” since I meant X and Y relation, which for analog PAL was 4/3 and for D1 is not 4/3 unless we change pixel aspect ratio, thats why 720/576*1.0667=1.333. With square pixels new digial PAL wouldn't be 4:3. that's the point.
In my example I was talkig about transfer film footage to TV. TV would assume non-square pixel and will stretch your beauty shot <–> which was on telecine scaned with square pixel assumtion. The way to keep geometry in shot is to narrow plate by 0.938086 specially that in case of 2K you have lots of resolution. Narrowing plate will cut off some pixels at top and bottom.
Even if you shoot on HD or any square pix format, but output for TV, this matod is applicable!
that's the point. I wrote “screen aspects” since I meant X and Y relation, which for analog PAL was 4/3 and for D1 is not 4/3 unless we change pixel aspect ratio, thats why 720/576*1.0667=1.333. With square pixels new digial PAL wouldn't be 4:3. that's the point.
In my example I was talkig about transfer film footage to TV. TV would assume non-square pixel and will stretch your beauty shot <–> which was on telecine scaned with square pixel assumtion. The way to keep geometry in shot is to narrow plate by 0.938086 specially that in case of 2K you have lots of resolution. Narrowing plate will cut off some pixels at top and bottom.
Even if you shoot on HD or any square pix format, but output for TV, this matod is applicable!
I think we're talking in circles. In the end, you need to render at whatever pixel aspect your background plate is at, regardless of what the various paths are(and there are many) between film, various HD's, and video. If it's SD video, then 0.9. If not, then 1.
I can't even tell if you're saying that you're disagreeing with me about something or not. You're talking about the various processes of how telecine gets a film transfer inside an SD frame(which isn't a Houdini process). There's different ways, which is typically determined by how the director framed the shot.
J.C.
In the end, you need to render at whatever pixel aspect your background plate is at, regardless of what the various paths are(and there are many) between film, various HD's, and video. If it's SD video, then 0.9. If not, then 1.I can't even tell if you're saying that you're disagreeing with me about something or not. You're talking about the various processes of how telecine gets a film transfer inside an SD frame(which isn't a Houdini process). There's different ways, which is typically determined by how the director framed the shot.
J.C.
yep, sorry, long day at the computer (18th hour right now actually…) :wink:
(“circle is a ideal shape” as Arystoteles said)
cheers,
SY.
(“circle is a ideal shape” as Arystoteles said)
cheers,
SY.
so all in all the final answer is….HD use PAR of 1? From what I've been working with it seems to make the most sense.
I'm really glad I stirred up this post…lots of good info
I'm really glad I stirred up this post…lots of good info
Yup - PAR of 1 for HD.
Cheers,
J.C.
Cheers,
J.C.
lynch_ppl
so all in all the final answer is….HD use PAR of 1? From what I've been working with it seems to make the most sense.
I'm really glad I stirred up this post…lots of good info
I personally have only encountered two situations where the pixel aspect ratio really is *not* 1.0: NTSC D1, 720x486 (PAR: 0.9), and Cinemascope, 1828x1556 (PAR: 0.5). Pretty much everything else has always been at PAR 1.0. I've encountered plenty of situations that I made up in my mind where PAR *wasn't* 1.0 but probably should have been!
– Antoine