#Pixel aspect discrepancy
26 messages in this thread
I was setting up my favorite resolutions in the 3DS.SET file and I
finally decided to ask about something that has really bugged me for
some time, but I just lived with it.
Why is there a discrepancy between the "RESolution" aspect ratios and the
standard formula for determining pixel aspect ratio.
Let's take the example for RES4 = 640,480,1.0:
480 / 640 * 1.3333 = 1.0 (accounting for rounding); correct!!!
This seems to be correct for determining pixel aspect ratio. Now in the next
example I will use an ATVista NTSC resolution:
486 / 756 * 1.3333 = .8571, but the pixel aspect ratio used by 3D Studio
for VISTAH is .88. If I want to set up the NULL DISPLAY resolution, which
resolution should I use?
And now for another known TARGA+ frame buffer discrepancy:
486 / 512 * 1.3333 = 1.2656, but the default 3DS.SET file reports 1.23
Even a old TARGA 32 has a pixel aspect ratio of 1.255 barely closer to 1.23:
482 / 512 * 1.3333 = 1.2551
D1 has a effective resolution of 720 x 486 and a pixel aspect ratio of
.9, but the NULL DISPLAY resolution in the 3DS.SET file reports .9091
for 720 x 486. Is there some special formula that 3D Studio uses?
I wanted to precompute all the resolutions I might need for any
particular delivery vehicle such as a PAR, but alas I am unsure of
the formula that I have used for years to correct any pixel aspect
ratio problems. Normally common knowledge would tell me the
following:
480 / 752 * 1.3333 = .851; close to .8571 for normal ATVista resolutions,
but as stated before, VISTAH reports .88 for pixel aspect ratio.
This has been annoying me for some time, so I was wondering what the
magic formula is that 3D Studio uses that is different from the 4 to
3 ratio that gives us 1.3333 of both TV and computer monitors?
Maybe I don't understand your question, but you have the formula right:
(Output Image Height / Output Image Width) * (4/3) = Aspect Ratio.
So, from the Render Configure dialog, just type in the desired W/H, do the
calculation for A/R, and type in the rounded number.
Am I missing something here ? You can 'load up' the dialog to default to
anything you want by changing the 3DS.set file.
BILL
Thanks for replying Bill,
I have always been curious about the defaults in the 3DS.SET file
because they always seemed to be slightly off from conventional
wisdom. I got another reply from Don Landis and if I am reading this
right, "Go with what you know!"
Bryce,
You're welcome…. << "Go with what you know!" >> … Good advice !
BILL
This pixel aspect thing has always been an area of confusion for me
and here's another reason why:
I'm preparing a homicide trial demonstration in which I need to show
the dynamics of the reflective property of an ellipse (as it relates
to a bullet wound.) I'm using a spotlight at different angles of
incidence against a flat surface to illustrate the relationship
between the light angle and the ellipse's major and minor axes.
So, I need the rendered circle and ellipsis to be accurate. But when I
render perfect circles at 640 x 480 using an aspect ratio of "1", the
circle I get does not appear to be symetrical — when I use a ruler
to measure it on the monitor. I do get a near perfect circle when I
use an aspect ratio of "99". It could be that I'm not measuring
accurately or that my monitor is weird.
Any thoughts on this?
Alec Jason
Alec, if you render to a square pixel and your circles aren't "circular" on a
640x480x1.0 display, then perhaps the monitor adjustment is off.
Do a test-rendering of a perfect circle, and display it. While measuring with
your ruler, re-adjust your monitor's X and Y axis image size settings. When
that circle is circular, lock it in and you should be all set!
When setting up a new system, this is usually the last thing I do after setting
up all the drivers (for 3DS and Windows).
-Alan
Check the vertical sizing and vertical linearity controls on the monitor. It
may need calibration for your atypical application. These adjustments are
usually inside on the chassis.
I've always been confused by aspect ratios, too. I certainly know what they
are; I can follow the derivations of course; but in the same way that it took
the computer video business several years to properly understand "gamma" and
color correction, I think there's similar slop in "aspect ratio". After all,
almost every monitor has width and height adjustments. Some TVs and video
monitors do, too.
So where do people get off quoting aspect ratios to four decimal places?
Hi John,
I enjoyed meeting you at SIGGRAPH. Well I guess four decimal places is a bit
extreme, but when I want to render to a Truevision ATVista with some 3D
software, that will go nameless here, and I want to use it for video and print,
the pixel aspect ratio comes into play. This is especially true if I don't
want to render it twice at different resolutions and pixel aspect ratios for
video and print.
The place where I do most of my work only has a MAC network and MAC printers,
so anything I print has to be downloaded via Farallon card to a MAC, massaged
and printed out to one of their MAC laser or color printers. Most of this
stuff just deals with square pixels.
Being "slightly anal" when I pull the image up in Adobe Photoshop, I want a
circle to be a circle so to speak. I also want to know that the circle is as
humanly and mechanically perfect as possible. I rarely want to depend on dumb
luck as I see some of these MAC people use around this facility. Personally I
feel that if I accurately stay on top of the technology that I am driving, I
will not get as sloppy as these people. I know most MAC people aren't as
sloppy as these people, but the impression that I get is that the software will
do everything for them including wiping their nose.
I also work with a DOS paint program, LUMENA, which only works with the
ATVista, but I also have Fractal Design's PAINTER for Windoze and often have to
go back and forth between them. When I want to adjust the horizontal scaling
(modifying the pixel aspect ratio) to correct the distortion cause by the
ATVista framebuffer under the square pixels of Windoze, it asks the operator to
input a scaling percentage to one decimal place. If I have accuracy to four
decimal places I can round to three, eg. 85.7% from a pixel aspect ratio of
0.8571. I can go back and forth a couple of times without accumulating too
much of an error.
Here's hoping my circle is as perfectly round as yours! <G>
Bryce W.
Oh, I agree completely – there's nothing wrong with being a stickler
for details… Except I always think of the final destination for
video, be it digital or analog. I cringe at the thought of how much
work goes into keeping video clean, when the final product is viewed
on a $100 VCR, and then there's those multimedia presentations that
get played in the corporate boardroom on a video projector that's
missing the 'red' signal, and no one notices… 🙂
Hi Alec,
As others have stated, it could be geometric distortion in your monitor for
various reasons. I'd take a look at the image on another monitor if possible.
Otherwise, I tend to believe in making any kind of adjustments as necessary to
please the eye, whether "correct" or not <g>.
BILL
Thanks, Bill, Alan, Don, and John;
I had a bad case of dopitis: I completely forgot about the height and width
adjustment controls sitting right in front of me on the monitor.
Thanks guys!
Alec Jason
You are correct and this subject was brought up two years ago. As for
what this 3DSer does its 1.266 for my Targa+ and 0.851 for the PAR.
Happy now? <G>
any news on the PAR front? You seem to have your finger on the pulse
🙂
1) New hard drive developements? (Anything faster than the 2110a/2217a)
2) What's the latest software for NTSC users?
3) Did you get the Razor Pro setup? (I'd guess not since you have the Fast
machine).
Thanks Don,
I was just wondering what the collective conscious thought about this anomaly.
I like your numbers and will use them, with your permission of course <G>.
Now I can sleep peacefully when I have time, with the knowledge I am "one",…
no "one-point-three-whatever-whatever-whatever" with the universe.
Bryce W.
CF/X Image, Inc.
>>I like your numbers
They're not my numbers as they come from equations like you listed. I recall
Greg Pyros explained the discrepancy as using figures from older targa boards
that used 480 or 482 vertical. Frankly, I believe all this is within tolerance
of the TV system anyway but if you have to insert a number, why not have it be
the correct one.
As others have said, Bryce, you've got the formula correct, and by
using it you should be OK.
It would seem some of the dicrepancies stem from simple mistakes to
small changes in the frame buffers' resolutions (say, from 480 to
486) over the years. I'm not really sure.
Re your question about rendering to NULL, remember that this setting
exists to match your rendering to the _frame buffer's_ pixel aspect
ratio. In other words, if your frame buffer or display device
displays a non-square pixel, then you need to render the image
accordingly. When you render to NULL, you'll get a file on disk–What
are you gonna display it on? If it's for print, or a square-pixel
device, then set an image size and a 1.0 pixel aspect ratio. If it's
for a display device that has a non-square pixel, use your formula
and you're all set.
Thanks Alan,
I am now three for three with "Go with what you know!" I guess some of us
don't like to work in a vacuum. Thanks for confirming my suspicion.
Enjoy your weekend!
Bryce W.
CF/X Image, Inc.
Anytime, Bryce.
-Alan
You have the formula right, and the discrepencies are usually
insignificant -unless you need to exactly match a previous
rendering's aspect ratio, etc.
A few points to keep in mind. A 4X3 format for video isn't always
exact depending on the framebuffer you are using. The PAR and the
ATVista DO have the same pixel aspect ratio, because they use the
same pixel frequency -the PAR just crops the image more. Video
displays 525 lines/sec regardless of whether your framebuffer
displays 486 or only 480. Calculating aspect ratio on a video device
is affected by how many pixels get crammed onto each line and whether
each line is completely filled or overfilled. The padding or
cropping of each line and the cropping of lines essential changes
your 4×3 value. Only your framebuffer's engineer knows for sure.
>>Video displays 525 lines/sec regardless of whether your framebuffer displays
486 or only 480.<<
Sorry but you are incorrect on this. 525 is the total number of horizontal
scan lines per frame on the video raster. This 525 (262.5 for each field)
includes the vertical blanking where no picture is present. The actual maximum
number of scan lines used for picture display on the raster is 504, on a targa+
it is further reduced to 486. The maximum allowed in the specs is 21 scan
lines for vertical blanking which reduces the picture area to a maximum of 504
lines. The specs refer to blanking, not picture display area. Most devices
further crop the picture area on the vertical by an additional 18 lines split
between the top and bottom of the picture ( this is not regulated). Using the
specification the maximum vertical resolution of NTSC video could only be 504
scan lines as the blanking would not count. Due to the way a TV picture is
drawn on the screen it would be impossible to cram more pixels into a scan line
than one so the maximum vertical resolution in pixels would be 504 pixels.
Computer devices use 480 pixels as one form of vertical resolution. It is a
simple matter to convert to one pixel per scan line than try to resample the
vertical information in the picture to 504/486 pixels per scan line, therefore
approximately 480 is used which includes a 7.5 IRE frame border. Depending on
your frame of reference this border may not be part of your resolution
calculation as it does not contain any picture information. As for overscan,
the amount is controlled by the receiver TV and is not regulated but if 10% it
would be displaying 525-(10% * 525) or 472.5 lines completely hiding a black
border of a 480 pixel display. The 504 figure is only theoretical as
"resolution" is an arbitrary figure and the generally accepted value is
luminance at 343 lines or about 70% ot numerical theory. This is due to
display screen capability and other factors such as electron beam angle
deflection refraction at screen edges etc.
The horizontal resolution is a completely different issue since it involves a
continuous trace and can be divided as much as the bandwidth of the video
signal allows.
Don:
I love it when you talk technical! More! More!
Greg Pyros
>>I love it when you talk technical! More! More!
Yea… and you're weird, too. <BG> (Or do I mean wired?.. spelling?)
Don:
>> Yea… and you're weird, too. <BG> (Or do I mean wired?)
Both! And proud of it! :^)
Greg Pyros
RE: Video resolution
Exactly! or, I hope you had that reply handy and didn't type it in on my
account. Actually, I was embarressed to see that I had miss-stated the
vertical frequency to be 525 lines/sec when it is 525 lines/frame -oops. I
think my point was: since no one makes framebuffers that display more or less
than 1 pixel/scan line, that the vertical cropping differences ( 480 vs 482 vs
486 ) create minor discrepencies in the calculation of aspect ratios ( the
question I was humblely addressing ):) In other words, calculating based on a
4×3 ratio will have small, usually insignificant variance from published
framebuffer specs which are based on the actual pixel frequency on each line.
Thanks for the reply, I'm not sure which part of my message that you didn't
like -just the incompleteness?
Paul Lempke -Rising Star Graphics
>>I'm not sure which part of my message that you didn't like
I just didn't want others to get the idea that the vertical picture was made up
of 525 lines when it is really 480 to 504. This can make a significant
difference when making the AR calculations. On the otherhand, I do agree that
the difference of 480 and 482 is insignificant. In Alec's situation,
documentation of the technique is also important and therefore he probably
needs to be mathematicly correct.