#Radiosity
29 messages in this thread
Actually, there is a Power Macintosh program that can do radiosity
(they call it "Raydiosity" which implies that it may be ray-tracing
with pseudo-radiosity) — Strata Studio. (It also has a mode called
Ray-Painting, which is ray-tracing with painterly effects — sort of
what you'd get using Pyros's Paint .IPXs or Penello.)
True radiosity is SOOO slow that even on the fastest supercomputers
in the world, it's impractical for animations (except for one special
case — see below), and is thus used only for stills.
The special case is architectural walkthroughs where NOTHING AT
ALL MOVES EXCEPT THE CAMERA. No escalators that run, no doors that
swing open as you walk through them, no mini-Venetian Blinds that
open, no Sun or other light sources moving, etc. The reason is that
radiosity's calculations are independent of the camera position, and
are a "computed solution" to extremely complex formulas based on the
contents of the scene (light sources and objects but not cameras).
So the calculations don't have to be done again after the first frame
so long as ONLY the camera moves.
Radiosity is only one factor in improving realism, and correcting
the "this is a computer generated image/animation" look that still
plagues all attempts at true photo-realistic rendering. Another,
which to the best of my knowledge NOBODY has addressed (not even in
the supercomputer realm, let alone SGI or Macs or PCs — Hey, Yost!
Here's something that r"5" can add to REALLY knock the socks off the
industry!) is TRUE PERSPECTIVE. All programs that I've seen use
Straight-Line Vanishing-Point Perspective, which is only an
APPROXIMATION for artists. Real-life perspective is CURVED (the
curves seem to be hyperbolas, but I may be mistaken on that). Proof?
Okay, imagine the typical perspective example of standing on a flat
plane in the middle of perfectly straight railroad tracks that extend
to the horizon in both directions. The tracks appear to converge to a
point on the horizon, but DO THEY DO SO AS STRAIGHT LINES? NO!! Why
not? Because they converge to points in BOTH DIRECTIONS! In front of
you AND BEHIND you as well! If the tracks appeared to be the
straight lines we know that they physically are, then they would have
to either DIVERGE in one direction, or MEET AT A DISTINCT (albeit
obtuse) ANGLE by your feet! And they DON'T! If you look straight down
at your feet, you can see that the tracks do indeed seem to CURVE,
being PARALLEL to each other ONLY AT THE NADIR (or other axial
line-of-sight)! The curve is most extreme at that point and quickly
becomes nearly but never precisely straight as distance from the
nadir increases.
You can also verify this by simply looking around your office, right
now. Look at a corner of the room. Look up as the corner meets the
ceiling, and down as it meets the floor. It appears to bend away from
you IN A CURVE in both of those directions, being closest to you as
it meets your horizontal line-of-sight. Look down a hallway and see
the same effect with the floor and ceiling. If your desk has straight
edges, you'll note the same effect with it.
Yet 3DS and ALL other 3D programs on ANY platform will render all
of these as STRAIGHT lines. You really notice it with animations!
Yost and ADesk, you guys got a REAL OPPORTUNITY here!
<<TRUE PERSPECTIVE Good explaination of the way we actually see with
curved perspective. Ive noted in my many years teaching color theory,
composition and visual analysis that most people cannot see the
curves. They know things like edges in architecture are straight, so
they see what they know rather than really observing their visual
field. In the same way the three point perspective that 3d programs
use is disconcerting to viewers who percieve the limited two point
perspective used in most architectual drawings as more realistic.
Render blowup in 3ds and special cameras for architecture "solve" the
problem of confronting viewers with a too real perspective of a
scene. I think that curved perspective would be great for giving
scenes the radical expressiveness of an Escher print but would not be
considered real looking by most viewers.
A work around is to map the animation onto a sphere, controlling
the strength of the effect by changing the # of degrees of the sphere
the map takes up. Ed
Actually, it's more in animation rather than stills that true
curved perspective is most important. Sure, in a still, most people
might not find the curves "realistic" unless they're very
understated. But watch an animated flythrough of a city or a building
in which the camera changes angles on multiple axes of view. There is
a very unrealistic look to the straight perspective then, which
curved perspective would help alleviate.
Quebillah:
Sorry, I had the need to give you a name, M&M just doesn't cut it for
me. <G> Anyway, have you checked out Lightscape for awesome radiosity
based renderings which are not sooo slow that one would need a
supercomputer? Maybe Lightscape is not true radiosity, I don't know,
the renderings are breathtaking.
Curved perspective rendering may sound like a wonderful idea but I
have my doubts that realism would be increased in the normal video
and print media. I suspect that an animation in VR could benefit
greatly from the effect, however. In print and perhaps in video, one
subconsciously compares all imagery to the image's or monitor's
orthoganal edges, frame, bezel, etc. Certain curved perspective
scenes may look worse as a result of this visual subliminal
comparison. Architectural images generated in 3DS and many renderers
often look threatening as buildings appear to be falling in to the
center of the scene as you tilt your camera upwards from the target.
I have one renderer, GDS, which allows one to employ 2 point
perspective which a lot of architects tend to prefer. In 2 point the
buildings are always vertical and line up with the edge of the paper,
a result which is very appealing in my line of work. Even though the
images generated in 2-point are distortions they are very
descriptive, understandable and visually stable.
To get a curved perspective view just add a little warp to your images
and your good to go.
>> Quebillah <<
Oh man.<g>
> true perspective <
Or parallax. A natural function of lenses. A good example of what
your referring to can be seen in the New City poster for R2 (and I
think you can render it off the R3 CD-ROM). In the image, the tops
of the buildings appear tilted and have acute corners — as if made
from parallelagrams. The rendering is intended to be a
high-perspecive view, but a "true perspective" image would look quite
different. For example, look through the fish-eye peep-hole in an
apartment door (a really high-perspective lens). The view looks as if
it was wrapped around the noze cone of a 747, or technically
speaking, a parabaloid.
I think the biggest benefit of a parallax-based system would be matching images
to various camera lenses for FX work. The drawback is a calculation overhead
caused by hyperbolic trig functions during rendering.
— James — Choreo Motion —
I'll have to check Strata Studio's radiosity abilities out. The other
thing to note about radiosity is that it's _always_ Gouraud shading.
That's why it tesselates the heck out of your model, so it can have
enough vertices where it needs them to provide the correct lighting
solution. Also, there are no specular relflections – every surface is
diffuse.
LightScape offers a somewhat unique solution by allowing you to
generate a second ray traced rendering ass to provide the specular
reflections. Of course it's only for a static view.
Now as for perspective <g>, you're of course correct – true
prespective is not perfectly straight. The question is whether an
observor could actually perceive the difference and if it's worth the
trouble in duplicating what is a subtle effect for all but the widest
lenses.
A correlation would be the phenomenon of color constancy, where the
human eye automatically adjusts for missing colors in the spectrum.
When observing a house at night its windows often appear "orange"
from its incadescent lights, yet when you enter the home it appears
"white" – your eye has adjusted for the lack of blue in the spectrum.
This is also a real world effect but one that the computer graphics
industry has taken a pass on for the simple reason that duplicating
it would cause confusion without providing a true benefit.
I have done architectural drawings on scores of enlarged photographs,
reverse engineering their perspective to derive horizons and
vanishing points, and I have to be honost – the flare you speak of is
_extremely_ subtle unless you dip below a 28mm lens. Granted, it's
there but you really have to be looking for it and need contrastingly
straight construction lines to actually see it.
Phil, not so. I wouldn't consider my eyes exceptionally trained (I'm not an
architect, engineer, or any other similar profession), yet even I'VE noticed
the unrealism in animations (NOT STILLS!!) caused by straight perspective. For
a long time, I didn't know what was causing the unrealism, only that it "felt
unreal." It was only when I analyzed what I was seeing that I figured it out.
But then, I'm good at that. I've read recently that there is still
no suitable explanation for why the sun and moon appear larger to us
near the horizon, but not larger to a camera. Some have claimed
"atmospheric magnification," but that would affect a camera as well.
The reason I have determined is that the sky does NOT appear as a
hemisphere to us, but rather as a quite flattened semi-elipsoid —
the zenith appears way closer to us than the horizon does. Go outside
and look up and then to the horizon and you'll see what I mean. What
we're seeing is the limits of the atmosphere above us, and that's
nowhere near a hemisphere of 180 but rather a much smaller portion of
a much larger sphere than we think we see — the tip of the spheroid,
as it were. Well, the sun and moon always appear about the same size
to us (approximately 1 second of arc in diameter), but when they're
low in the sky they appear much further away since the sky itself
seems further away, yet are the same size when the rules of
perspective dictate that they should appear much smaller since
they're further away. But a camera is not 3D and so does not see this
apparent difference in the distance in the sky based on ascension.
You can verify this by staring at a rising or setting moon that is
very close to the horizon and thus looks quite large, for long enough
to form an afterimage on your retina. Then look up, and you'll see
that the afterimage appears the normal size of the moon as it would
be at that height in the sky, because the same size image is now
projected by your retina onto a sky that appears much closer than at
the horizon.
Moral: for the most in realism, don't map CLDMAP1.TGA to a
hemisphere, or if you do, flatten it considerably (2D Scale it
vertically to about 33%). Of course, link it to your camera so that
the zenith always remains overhead, but also Link Info it to remove
the rotational linkages.
Interesting discussion on perspective distortions, but I have never
heard anyone actual comment…
"Nice image, but….the perspectives should be more bent over in this
left corner"
The fact is that not many people are even aware that this
happens(myself included until your comments) means that no one has
ever found it strange that they are not there. I tend to think that
the effect you mention is more a result of optical distortions
involving our physical eye rather than mathematically accurate. If
this were the case then it would indicate that the light coming from
the railway line was actually being bent, and although railway lines
are heavy they don't exert that great a gravitional force in order to
bend light <g>.
The human visual system is not perfect and some inherent defects means
that you'll get some distortion. Seeing how a fish eye lens distorts
(traps) light, it is easy to image that our own eyes would also
distort incoming light, albeit less.
Maybe you could send me an image of with hyperbolic perspective
distortions and I could see for myself.
DaviD "Render/Render View/Bendy" GouID
<< I tend to think that the effect you mention is more a result of optical
distortions involving our physical eye rather than mathematically accurate.
I think its inherent in the scene. If one were to hold a yardstick in
front of ones eyes, perpendicular to the line of sight, a few inches
away, the stick would appear wider in the center because its closer
to the eyes than the the ends which appear smaller because they are
further away. This is not to say that the eye dosnt cause some
distortion as it is a wide angle lens.\
David, you miss the point. Of course light does not bend. But we're
not talking about light. We're talking about angles of view. Ever see
string art? You know, made of straight pieces of string tied between
pegs, but the result looks like it has lots of hyperbolic curves even
though it's all formed from straight lines? This is a similar effect.
The line of sight from our eyes to any point in our FOV is, of
course, straight. Likewise for a camera. But the combination of those
straight lines cause hyperbolic curved intersections resulting in
curved perspective.
I don't have to show you a picture. Just go stand in a long, narrow
hallway and look at where the walls meet the floor and ceiling and
see for yourself. Yes, it's an optical illusion — ALL perspective
is an optical illusion (no, those trees on the horizon aren't REALLY
the size of ants!). But it's an optical illusion that needs to be
duplicated by 3D animation programs to make for more realism.
As for people not noticing WHY an image seems unreal, see my reply
to Phil on the subject. Most people don't know the technical terms,
or could put into words why something doesn't look right. The name of
this thread is "Radiosity." No one here would deny that radiosity,
real or simulated, greatly enhances the realism of an image. Yet
would the average person know that the lack of radiosity in an
otherwise very realistic ray-traced image is why it doesn't look
quite "real" to them? Either by the term "radiosity" or even being
able to say, "Uh, shouldn't that yellow pencil be reflecting a bit of
yellow glow onto that wood desk?" Of course not. They just know that
something doesn't look quite right, but they don't know what. So it
is with curved perspective.
I enjoyed catching up on this thread and a have a few comments -FWIW.
1, 2, & 3 point perspective are manual construction techniques that
don't bear much resemblance to the algebraic perspective method used
by typical 3D renderers wherein a point in space is projected onto a
frame standing between the object and the viewer.
If you hold an empty frame up in front of you and look down the hall,
you will see that straight lines really are straight. The frame that
we view graphics through determines a lot of how we percieve these
things and since the frame is itself an external object in our field
of view our perception of it is subject to illusion as well.
If the image is large enough to immerse yourself in, then the
image/object will exhibit the same distortions that you mentioned -it
is a matter of scale and reference. One cheap way to distort your
images is to cut a matte that bows in slightly all around -the matte
will look straight and the image will have the curvature that you are
looking for. This essentially corrects for the disparity in scale
between the scene and the size of the rendering. ( Ever notice how
Zenith flat CRT's always appear to bow in?)
Hand held prints of small objects in a shallow space should need no
predistortion. Similarly wide screen movies of architectural spaces
shouldn't either, but when you squeeze a building into a TV set then
you might want to. An image processing routine (IXP) should work
fine, I bet Mirage would do it.
You're just not getting it, Paul. Sure, the straight lines would
line up with the straight lines of the frame, but that's because the
frame itself is also subject to perspective and is thus visually
curved to the same degree. Besides, the main point here is in FLYING
ANIMATIONS where the camera, and thus the POV, is CONSTANTLY
CHANGING. It is here that the inadequacies of straight-line
perspective really show up. Look for instance at some of the
animations in the 3DS SIGgraph Demo Reels of the past few years: the
architectural ones in particular show this effect. When the camera is
making sweeping moves around geometry, you see a sort of distortion
in the geometry caused by the straight-line perspective that might
not be noticeable, and in fact might even be desirable, in a still,
but is UNdesirable when the camera is moving! This effect is hard to
describe in words, but it's quite apparent if you know what you're
looking for. If not, you just might notice that it doesn't look
quite "right" without knowing why, much as someone who isn't a
typographer may not notice WHY an ad typeset by a neophyte with
PageMaker looks wrong when it has " and ' quotes instead of true open
and close quotes, no kerning, mal-formed fractions, mis-matched
fonts, etc. — they only know that it DOES look wrong.
This is interesting. Given that the frame itself is subject to the
same perspective curvature — as might be everything we see —
shouldn't it follow that the "straight" lines in a computer graphic
image are also subject to the same curvature? We are, after all,
viewing a 2D projection through the same eyes that apply curvature to
nature. In addition, it seems like the perception of curvature is
only available by comparison (with what we might call non-curved, or
"straight" objects). In nature, we don't really notice curvature,
partly because there's nothing to compare it to. That is, what we see
in life is not usually enclosed in a frame to cut off our peripheral
vision (where the curvature is more noticeable), whereas computer
graphics are always enclosed in an artificial frame (monitor or movie
screen), which might make a "naturally" curved image seem unnatural
by the very nature of its unnatural frame.
Sorry, just had to get that out of my system. <g>
– Jack
Jack, in a word, no. The screens we are viewing are flat (or very
nearly so). They certainly don't have the real-world depth that
their images portray. So they aren't affected by the curvature of
perspective. And that is, in fact, the problem.
As for viewing through a frame, if the frame is tall or wide enough
to enclose, say, our view of a hallway, then it would extend far
enough out of the horizon plane of view of our eyes to be affected by
the curvature to the same degree as vertical walls and corners in the
hallway would.
And you're correct that we normally don't notice this because of a
lack of reference. But remember, what I'm talking about here isn't
in stills where the difference would matter little it at all. I'm
talking about moving-camera animations where the difference becomes
noticeable as a "computeriness" or "artificialness" that we see in
the animation, often without knowing why, despite every other aspect
being as realistic as possible (radiosity, photo-realistic materials,
motion-blurring, etc.). I first started wondering this when I myself
noticed the effect in many 3DS SIGgraph demo reel animations,
particularly architectural flyarounds and flythroughs. It took me
awhile to figure out just what was wrong, but I did know that
_something_ was wrong. Finally I figured it out by comparing the
perspective in 3D images vs. that in photographs taken by a camera,
and by doing real-life observations using my own eyes and just being
in a room or hallway which should have straight lines but which
appeared curved.
>>…using my own eyes and just being in a room or hallway which should have
straight lines but which appeared curved.<<
Do you, by any chance, wear glasses (for nearsightedness)? I say that
only half facetiously, because I myself usually wear contact lenses.
While wearing the lenses, I don't really notice much curvature, even
in my peripheral vision. But when I put my glasses on at night,
suddenly all the peripheral walls are curved. In addition, our visual
perception is not based on a fixed view, but on instantaneous,
flickering glances from spot to spot… We don't really pan like a
camera, but jump from focus to focus and then generate a gestalt for
our brain. Just because a camera displays curvature doesn't mean that
our eyes do. And even if our eye display it, that doesn't mean that
our brain expects it (unless our brain has been conditioned to relate
"reality" with what the camera can translate). Regardless, I agree
that it would probably be nice to be able to imitate in 3D graphics
the curvature effect of the camera lens.
– J
Hello J
<<Just because a camera displays curvature doesn't mean that our eyes do.
As Escher pointed out curved perspective is not the result of lens
distortion but is due to the inescapable fact that we view things
from one point in space. For example if one is in front of a wall
looking straight at it, the ends of the wall will appear smaller than
the section directly in front of us because they are further away.
Big in the middle, smaller on the ends, it has to curve.
<<And even if our eye display it, that doesn't mean that our brain
expects it
I agree with you, most people see conditioned by what they know, that
lines in architecture are straight. The greeks actually corrected for
these effects and others in some of their architecture. The
Parthenon contains some very subtle curves to create the appearance
of straightness. They must not have had a tight budget.
Good points, Ed. I should emphasize that I'm not saying that the
capability of reproducing curved perspective wouldn't be a good thing
to have — I just suspect that it wouldn't look "real" to the viewer.
On the other hand, used properly, you could probably achieve some
"softer" looking imagery. Used at its extremes, you could get a nice
cross between German expressionism and Toon Town.
– J
<<suspect that it wouldn't look "real" to the viewer.>> Well, that's
one of the great things about art-making. Taking the viewer away from
habituation, rather than dedicate tons of equipment and technology
proving one can fascimilate r-r-r-reality. I've been playing with
curved perspective by taking portions of a scene and mapping to
subtle (and some times not so subtle spheres), or running TGAs
through an Fx which allows selection of an area (over time, animated
if need be) producing a curved perspective look. Used in combination
with crisp focus against blur and pulling the blur with an increase in
curved perspective. Also very effective when accentuating POV of a
character. Jeffrey
Jack, it would certainly be possible to imitate any lens curvature
effect with an IXP. All it would take is an understanding of
optics. I'll bet some folks could even do it in 5-10 minutes. <g>
>>certainly be possible to imitate any lens curvature effect with an IXP
I agree, but think how much fun could be had by opening up the
camera! You may remember that I've always wanted the capability to
translate the camera viewport for creating stereograms, multiple
channel video installations and multi-image slide shows, maybe there
are other transformations that could be linked to the camera view?
<< lens curvature effect with an IXP >> Yupperoooo! A great optional
camera choice, with varying degrees of lens curvature [18mm fisheye
as 7' barracuda actually notices …..gurggle…breathe….gurggle
ooodon't moveooo. I've been staying away from scripting much. If I
get comfortable there it's possible to never come back. Writing IPAS'
would certainly make me captive. The other day was working with some
of Jonas' Keyframer ideas in scripting… really nice, elegant
[SOMEBODY STOP I'll end up sitting parallel to the ground, head
inclined, with that non-communicative programmers smirk on my tinted
lenses, Bose speakers blarring — or does that only happen in UNIX?]
Jeffrey
<<Given that the frame itself is subject to the same perspective
curvature — as might be everything we see — shouldn't it follow
that the "straight" lines in a computer graphic image are also
subject to the same curvature? <<In nature, we don't really notice
curvature, partly because there's nothing to compare it to.
Another factor is that we view computer screens, tv screens, prints
etc. in the middle of our visual field where the curve is less
extreme and therefor less apparent. In real life the view includes
our perifery where the the curve is more extreme and apparent if we
see it. One reason most folks cant see the curve is that they dont
pay much attention to their whole visual field but focus in the fovea
macula area in the center. When the curves we see on the edges of
our vision are recreated and then played back to the center of
vision, fish eye "distortion" occures. On the other hand the same
image might look right in IMAX which fills the whole visual field.
>>whereas computer graphics are always enclosed in an artificial
frame
Until we have total immersion, that 's true and important for any
computer artist to keep in mind. Although it needn't always be
'artificial'. If you truly use the frame in your work then in
becomes neccesary, usefull. For instance, I've done a fair amount of
work in holographic stereograms that have ended up as hand-held
objects de art. In this case the perspective space of the 3D scene
is best linked to the frame so that as the viewer tilts the frame
side to side -the 3D objects follow. This of course is a totally
different senario from an architectural walkthru where the viewer
tries to immerse himself in the scene, but I think a projection large
enough to immerse yourself at scale would display the appropriate
visual cues.
I still like the idea of distorting the frame to create a
compensating perceptive distortion.<G>
Sorry to rekindle a thread and disappear, I've been unexpectedly
busy.
>>Sure, the straight lines would line up with the straight lines of
the frame, but that's because the frame itself is also subject to
perspective and is thus visually curved to the same degree.
Exactly my point. I think that Jack sees the significance of this.
Amoung other points I wanted to make was to complain about the use of
the term "true perspective". 3D perspective projections are the real
thing -which is no small thing. In the Renaisance when the 'Queen of
all sciences' was invented, lifetimes were devoted to competetions
for techniques that best approximated the 'true perspective'
projection that we only now can take for granted. Be that as it may,
distortions are often necessary for effect or to compensate for the
fact that you are trying to squeeze a panoramic view into a too
little frame. In our business there is always a 'frame of reference'
<G>. BTW -perspective was defined explicitly in terms of a frame.
Apart from that, I suspect that you are also comparing animations to
live camera scenes with their inherent optical distortions. Funny
how movies become our standard for reality<BG>. In either case, I
still think that an IXP would do the trick. When I was in graduate
school, I wrote a several programs that performed image manipulations
similar to what Mirage does. I think Mirage probably would even do
the job here. I enjoyed your postings, I appreciate thoughtfull and
thought provoking threads. And yes, I do get it.
-Paul
There's a radiosity company out there called LightScape, founded by
the guy who started Autodesk's multimedia division. They use
ordinary 🙂 SGIs to calculate the scenes, then perform real-time
walk-throughs on the scene.
A couple of quick corrections:
1) Lightscape was not founded by that ex-Autodesker. It was a fully-
functioning company with completed code well before he came on
the scene.
2) David Kalish started Autodesk's multimedia division, and David
hired the LS guy to take over from him.
– G
Sorry about that mistake; an Autodesk employee introduced me to that
ex-Autodesker in that way at the NAB party. Yes, the liquor had
started to flow before the introduction, although I hadn't any yet.
🙂
I'm just trying to stay on top of "revisionist" history. (no prob)
– G