Question About Realworld Lighting Simulation!

[user_deleted]

Guys :

I got some questions to ask.

recently, I am doing some experiments on lighting simulation for my thesis.

My question is how to exactly simulate the color of the light from the real world ?

The purpose of doing this test is to achieve a realistic composition later on, but I think the first thing is to match the daylight color to the overall environment color, such as color from the sky in the middle day.

could someone show me one of your ideas ( or your experience ) to achieve a satisfactory result in this stage ?

Big thanks, mates.

[joojaa]

buy a good camera, then stich up a hdr lightprobbe. Or just buy existing ones. This will give you measurements of teh sky. Then its relatively easy to form a scene that works almost right.

The problem with exactly right is that, reality has infinite fidelity. And to simulate real light yoru light probe should ork on etchnology taht does not exist for camers or monitors. Youd need to record the spectrum and calculate bounces witha ll spectral data. Problem here is it gets SLOW, so slow its hard to imagine anything taking so long.

Second threes no such thing as average daylight, each position of the globe is different form each other and each day has slightly different intensities on all spectra. Just like theres no average wood texture, no wood is this so it would look off.

The daylight gizmo in maya is pretty deecent at doing teh average tough. But because of this its not good enough.

[user_deleted]

Thanks for answering.

I know there is a method for simulating real world light colors with HDRI.

But, with that exception, is any method to extract colors from the real world images and apply them to 3D virtual lights to simulate a realistic color of lighting environment ?

Any idea ?

[joojaa]

For most practical uses a good digital camera is a very exact measurement instrument (certainly the most usefull one). But you can also just buy a photometer, or a spectrometer.

But how to measure is a complex issue, not really in the scope of theese boards. Go to yoru uni and ask for lecture about this.

QUOTE(zhimama @ 01/28/09, 02:09 PM) [snapback]299796[/snapback]
I know there is a method for simulating real world light colors with HDRI.

Actually hdri is not necceserily a striict measurement f intensity unless you happen to know awfully lot about the data. But turns out only the relative intensities matter for anything.

Well you can downboad light profiles wich are based on the measurements. Measuring light is made with a photodiode nowdays. Guess where you will find the meost accessible fotodiodes? Yes its in your camera so you would use yoru camera for measurement.

However if your actually wanting the spectrum of your measurement then you need a spectrometer. (not to be confused with mass spectrometer, well it contains same sensor principle)

QUOTE
method to extract colors

Define color. See many different spectra will result in humans seeing same color. However when they hid different colored objects they behave differently.

So extract in sense rgb+ intensity or spectum of intensity? If spectrum of intensity what resolution of spectra? What resolution of intensity? If you want spectral then this brings the question are you prepared to write ALL your shaders a new? Then how do you want teh data output, monitors dont exactly show measureble intensities accurately, nor does print.

See the problem is that our renderes do not work like reality so theres not much point in exact measurements. Because having exact measurement does not help without exact surface materials. All the basic shaders are more or less just approximations, or like in case of lambert something taht ddoes not really exist.

Scientific measurement is also a bit hard, i mean scientifically exact measurment works like this: Get a measurement instument, read its instructions wirte a paper on how it works. Then set up a experiment measure several times, plot using different data ranges to obtain error metric. Estimate error of your measurement. Compare to others results. Eliminate noise start over again and then MAYBE you have something halfway deecent. Then you end up with something thats very very little off form exact,

PS: im not the one whos nitpicky here im telling how measurement actually works. Its not s imple buy sensor thing, on top of sensor you need to have pretty good picture how the sensor works to understand its output.

PPS: how to make a perfect measurment is kindof something they are working out the since forever. It the kind of thing thet takes years to make work

[user_deleted]

Thanks for reply quickly.

The reason I am trying to do this measurement is that I had experience to teach 2-3 grade BA students in digital compositing lesson. I gave them a series of rendered sequence images( Foreground images were 3D rendered elements, BG images were HD sequences ) and let them to composite them into a seamless composition, which has to do some serious color correction. And the results I got from them were so different. ( 30 students in this class, but I received almost 30 different final results )
I think if there are some basic disciplines to follow, not just modify them by your artistic sense, it might be easier to obtain a closer result. Same thing in 3D part, if we can measure ( or extract some reference colors from live footage images, then apply these colors to the virtual lights ) it, then might be easier to get a closer result with the live footage, which we intend to composite with. ) Actually, it is very important part in this stage.
We all know that in order to achieve a final realistic result, ( 3D elements with live footage ) ,we have to match the color of objects( light colors/environment / reflection colors ..) in 3D space as close as possible, then render them out for the final composition in compositing applications.

[joojaa]

Just out of curiosity wouldn't it have been easier it make a light rig and then force everybody to use it?

well you asked for it, here comes. Point is taht doing teh convokluted match si the easy way out.

QUOTE(zhimama @ 01/28/09, 03:14 PM) [snapback]299801[/snapback]
I got from them were so different. ( 30 students in this class, but I received almost 30 different final results )

Yes thats expected. Did you actually have a color profiled monitor setup? Did you tell them to do so.

Also yes 30 persons all have different visual systems that work differently from each other.
QUOTE
The reason I am trying to do this measurement is that I had experience to teach 2-3 grade BA students in digital compositing lesson.

Ive actually done this too, and the subject is way too advanced for these poor souls They can not be made to agree period.

QUOTE
We all know that in order to achieve a final realistic result, ( 3D elements with live footage ) ,we have to match the color of objects( light colors/environment / reflection colors ..) in 3D space as close as possible, then render them out for the final composition in compositing applications.

Ahh its actually not that simple. Obtaining the color is just one step on the way. If you obtain a rather exact rgb value of a objects color then it brings out 2 additional questions?

What metamer did you get?

What does the rgb value mean?

It turns out BOTH these questions have a HIGHLY complex technical answer.

When you measure a olor in real world and translate that into a 3 component color you loose a infinite amount of information. The reason we use rgb color is that our eyes actually work in this manner, it quite literary has 3 sensors a red green and a blue one. They all see a range of different wavelengths and report out a singnal. But see nothing is fo this color in actuality. What happens is that each object emits bands of color spectra.

A argb color of 0022FF (see below for what rgb colors are), is actual sane solution for MANY different measurable light values. So even if you measured the color of a object A to be something and a object B to be exact same rgb value does NOT mean they will behave exactly same in different light conditions.

As a example i can take my heavy duty winter coat that's bright emergency orange (i want to be abale to see myself in satellite picture, or seen by cars )I can actually measure the color quite exactly with a camera. But i only get on metamer representation fo it. It turns out this jacket has a very special spectrographic color. Where it is a continuous red, but with a VERY thinn sharp spike of yellowish green that tilts the perception into emergency orange when in daylight. But lo behold cut of a very small piece of spectra, and my jacket becomes bright red even tough all other colors are roughly the same.

So measuring the value is simply not possible with standard shading models. The real world materials dont work exactly this way. This brings us to:

QUOTE
or extract some reference colors from live footage images

Yes that would be what the color picker is there for. you don't actually need the exact real value you just need to reproduce that exact shade of rgb color to match. But if you want each of them to reproduce something that is comparable you actually need to inform them of the palette beforehand. S o they can agree on it.

Now back to the rant

What is rgb color, well it turns out is not simple. Each monitor produces a different color for same rgb values. So if you didn't have color correct calibrated monitors you'd have different looks, even slight changes would make same thing. So did it occur to compare the footage on their original monitors in their work conditions? This would probably account for the bulk of the difference.

So if you want reproducible exact same colors you also need a color matcher like this one:

QUOTE
http://www.dpreview.com/reviews/colorvisionmonitorspyder/

And every one in the loop including the camera need to be calibrated , and profiled on regular basis at least once a month. Each workers station should be calibrated and profiled either continuously or set once so lightning conditions do not vary.

Now this gives you map to what they were seeing when they did their work. We call theese color profiles. But the profile only helps if the client has a monitor capable of same colors. (more on this later)

Alas we are not done yet the renderer assumes your working linear, so all the colors you measure need to be worked out to linear space before entered in the renderer. So the user will not see same values in swathes as he gets in output. This is a important step or your color will render out with wrong intensities when combined with global lightning. You then need to reverse this to get into your output.

Now before i tell you about why color profiles don't help squat, i will tell you what color profiles DO when they fail to match. They do one of 4 things: Clip the colors and just show the closest possible equivalent, clip but with special twist so that white seems to be equalized, choose nicest bright color, Scale your color space to match the one your in and only show the relative colors. The latest is used because as it happens to correspond to the way humans are sensitive to color, and automatically max range everything within reasonable tolerances.

So now what really matters is that you get a color value that is relative to the image in question, not any absolute measurements. Basically because there's so much on the way that can go wrong. So what you doe is you make a up a palette beforehand and say is am for this color palette. This way all artists know how to get consistent look. But before thet they have to have good monitors that are all roughly same capable, color profiled regularly and color corrected. Using the same approaches and shader libraries..

Now once you have done that ill tell you what will go wrong. See you now have and don't have:

Correct color values measured

Incorrect representation fo the real color because you have tinted metamer

correct profiles

correct render pipeline

correct output and input

comparasion sheets

So now what is the problem? Well your customer, he doesn't have a color correct output unless your extremely lucky. And even so he will only see the relative colors. So at the end of the day while this may help in studio the only step thet you really need is the color picker one, because thats the kind of level your output sees.

And all this si destroyed by auto white point in the end.

Well this is the gross simplification!

But you measure light color like this put a white paper in scene, then remove the universe and just leave the paper an light*, then measure distance form light and viola the color needed to produce whatever match you want. But yes you don't need this step your shader td already can arrange the situation so it doesn't matter. Alternatovely do the gray ball thing below.

• see easy.

QUOTE
in 3D space as close as possible, then render them out for the final composition in compositing applications.

No its not that simple. Youd actually need a guy on the set that measures everything. 99% of every shot dont have that luxury exacly because they didnt anticipate this need in advance. Thats why we have color correctors. They are cheaper.

So you see doing the convoluted color correction IS the easy way out. Its work but infinitely less so than foing around measuring sets exactly. So new yu can amke this easier but then youd make everything esle much harder. Its just easier to do it here.

PS: there's a nifty trick called gray ball sampling where you do a hdri of a gray diffuse ball and use theat as lookup for lamber colors. It doesn't accounts shadows right but on the whole sahdes mighty right.

PPS: so if you had them do renders without very intricate knowledge on how cg works its GUARANTEED you got it off by mile. Wasn't it you who was supposed to teach them all this. I guess not, to this day and age it common that only ONE person in the entire rendering house knows this and is kept busy trying to void it alltogether