Welcome, Guest. Please Login or Register
Home Help Search Login Register
Pages: 1 2 
Send Topic Print
Unclipped hdr (Read 11682 times)
brainspoon
Ex Member


Unclipped hdr
05/18/12 at 00:56:46
 
Anyone who knows a good way to shoot unclipped hdr with bright sun light?

Because of this paper I was asked if I can do this:
http://www.fmx.de/program/event-details/event/389/610/45/610/detail/Event.html

The only way I can imagine to shoot such is by using a full format camera with a lens where you can put a ND filter on top.

I believe quite some time ago some students where testing this by shooting a nonlinear sequence of raw files. A few very dark esposed for the sun and then the normal exposures.
Back to top
 
 
IP Logged
 
Blochi
Administrator
*****
Offline



Posts: 1726
Hollywood
Re: Unclipped hdr
Reply #1 - 05/18/12 at 02:59:46
 
Interesting article.

I really think it is debatable, and I can see how one can talk about it for a full hour.

Unclipped lights are critical when you're shooting the lights and spots on set. Even though they are very bright, they can typically be captured just by aligning the exposure brackets correctly. They are not as outrageously bright as the sun.

The sun is special.

For several reasons I stopped worrying about the sun:

- You're right - it does take some extra ND filters.

- It sets you up with a complex EXIF manipulation step, because these extra exposures would otherwise not get registered correctly. Okay, that just got easier with the new Lightroom4, but still...

- ND filters do not fit on a fisheye. (there are a few that have a filter holder on the lens end, but most fisheyes don't and are just too bulky glass for a front filter thread.) With a longer lens (where a filter would fit) the shoot takes longer, to the point where it is impractical on set.

- And then: for what? There's no mystery what to expect from the sun. It's a circle. It's bright. You can also just paint it in. Actually, I have a new tutorial in the upcoming book about that, this picture should explain it all:

...

Of course, I don't encourage to leave the sky or immediate surrounding of the sun blown out. You should at least shoot enough exposures into the highlights until the circle of the sun is clearly visible. But color and intensity you can boost yourself, and that actually gives you some artistic control over it.

That is really the final argument that made me stop attempting to capture the full DR of the sun: It's just such an important factor of the lighting. As someone who also uses his own HDRIs to light a scene, I really do prefer having control over my key light. That's why the sun is specially tagged in sIBL files, so it becomes a regular light source where I can adjust sharpness, color, intensity based on creative decisions and client feedback.

So.... for strictly unbiased render engines the sun would need to be painted in. Renderers like Maxwell or Keyshot, that really only rely on the information in the HDRI. For normal production renders, it's better to use an adjustable light source to supplement the HDRI. Since that comes in as addition, you would have to clip the sun in an HDRI anyway - otherwise you would have double the sunlight.

Long story. But I hope it all makes sense.
Back to top
 
WWW  
IP Logged
 
brainspoon
Ex Member


Re: Unclipped hdr
Reply #2 - 05/18/12 at 09:27:03
 
I use the painting technique from time to time too. I just pick a color close to the sun where it gradually fades away.

Maybe as an experiment I will try something like this
http://www.magic-filters.com/tokina.html
Back to top
 
 
IP Logged
 
Blochi
Administrator
*****
Offline



Posts: 1726
Hollywood
Re: Unclipped hdr
Reply #3 - 05/18/12 at 09:40:37
 
Let me know how that goes ... very curious.
A full-on experience report would also be a great guest blog post Wink
Back to top
 
WWW  
IP Logged
 
brainspoon
Ex Member


Re: Unclipped hdr
Reply #4 - 05/18/12 at 09:45:18
 
Hehe. It is too bad that I have the Tokina 10-17mm lens. There seem to be quite a few others that have a holder for filter gels.
Back to top
 
 
IP Logged
 
davidb
Ex Member


Re: Unclipped hdr
Reply #5 - 05/18/12 at 10:39:23
 
Blochi wrote on 05/18/12 at 02:59:46:
Interesting article.

So.... for strictly unbiased render engines the sun would need to be painted in.


Just to be pedantic.  Whether or not a monte carlo renderer is biased or unbiased is more to do with whether or not it is guaranteed to converge on the correct solution than support for direct lighting.

It is quite possible to build something that is unbiased and handles a sun light + sky perfectly (bidirectional path tracing, metropolis light transport, hybrid path tracing + direct lighting etc.). 

And it is also possible to write a very biased renderer that uses image based lighting for everything.  Anything that uses an irradiance cache would qualify here.

The fact that some of the best IBL renderers happen to be unbiased is merely coincidental Smiley

/pedant
Back to top
 
 
IP Logged
 
Blochi
Administrator
*****
Offline



Posts: 1726
Hollywood
Re: Unclipped hdr
Reply #6 - 05/18/12 at 17:08:01
 
Well, point taken, all those things are theoretically possible. And I think many unbiased renderers actually do have a special "sun" object, which only underlines the fact that the sun is of particular interest and does not necessarily belong into the HDRI.

Could also build my reasoning the other way round: Whatever you want to call your renderer, all the HDR tricks are powered by Global Illumination algorithms. That's where the ambient and bounce light from the scene comes in handy. But the sun is a prime example for a Direct Illumination light source. Which is quite the opposite. Even though it's not impossible to shoehorn this through a GI algorithm, and some of them nowadays handle it gracefully with importance sampling, this is simply not its native field.
Back to top
 
WWW  
IP Logged
 
davidb
Ex Member


Re: Unclipped hdr
Reply #7 - 05/18/12 at 17:44:07
 
Agreed.  It's hard to get right, and importance sampling just steals sampling (and thus cpu time) from the rest of the sky.

Why take fire hundreds of rays ay the sun when you get almost the same result from one?
Back to top
 
 
IP Logged
 
Gerardo
Senior Member
****
Offline



Posts: 74
Re: Unclipped hdr
Reply #8 - 05/23/12 at 04:18:18
 
Adding other issue to Blochi's list is that many of those "neutral" high density gelatins that you put between the fish-eye lens and the camera are rarely chromatically neutral, they commonly introduce color variations than later you need to profile and color correct with a matrix/LUT. 
 
You'll need additionally other bracketing sequence (besides the 16-18 f-stops that commonly cover the skylight) by changing not only the shutter speed, but also the f-stop (let's say f/4 - 1s, 1/4s, 1/32s; f/16 - 1/16s, 1/125s, 1/1000s, 1/8000s...) to be able to capture the whole dynamic range - at 2EVs spans it would be about 2^22. 
 
Some people are more picky about this than others but having the full values range of a scene can be useful for replacing later the sun area with a 3D light. Perhaps other approach (in cases where full HDR capture is not possible in location) might be linking the light position to a physical sunspot (latitude/longitude) and use a RGB modifier in sun's color that takes the values from a sunsky texture environment.



Gerardo
Back to top
« Last Edit: 05/24/12 at 04:26:25 by Gerardo »  
 
IP Logged
 
brainspoon
Ex Member


Re: Unclipped hdr
Reply #9 - 05/23/12 at 12:12:28
 
http://www.hdri-hub.com/freesamples/Unclipped_Test1.zip

A first quick test. Not unclipped, but I am getting closer. There is much room for improvement.
Just lighting with the hdr gives pretty good shadows.

A nice trick I found yesterday is, in vray you can easily map the environment into an area light texture and then use that for lighting.

Thats what they do in some movies now. Cut out the lights from the hdr and map them to area lights to get proper positioning in space. Also saves some samples during rendering.
Back to top
 
 
IP Logged
 
jasonhuang1115
Ex Member


Re: Unclipped hdr
Reply #10 - 05/23/12 at 22:26:37
 
As my understanding, the "unclipped", in this particular talk, is referring to half-float or full-float format of the HDR env. maps vs. 8-bit clipped maps. In the past, there was no importance sampling for raytracing in renderman, so they used "clipped" env. map (8-bit) for diffuse lighting. With recent releases of PRMan, they are able to directly raytrace the HDR maps that have hot concentrated light sources.....

Paul Debevec had done capturing full dynamic range of a scene with the sun visible in the scene in 2004. I agree with Bloch that it's more flexible and sampling friendly to use a CG light to represent the sun and calibrate the color of the light and map via color chart and/or gray ball.
Back to top
 
 
IP Logged
 
Gerardo
Senior Member
****
Offline



Posts: 74
Re: Unclipped hdr
Reply #11 - 05/24/12 at 04:29:18
 
Brainspoon, nice test! Here the HDR map seems to cover a bit more of 19 EVs. So it's pretty close.

The sampling issue begin to show up in my renderer. Diminishing the environment map size and applying an HDR filter help. Without touching the size, the GI interpolation algorithm also help here. Agreed about the trick you describe. in LightWave we have also a couple of nodal lights for mapping images (from area lights/geometries). Studio lights can be mapped also in this way like the HDR-cookies described by Blochi in his book. Before these type of lights, I used panels rigs based on the spinning light trick. It's indeed a useful trick mostly for IBL in indoors.

Btw, guess the sIBL-edit could be useful for getting the sun values in a common light. 
 
Jason, I think it's referring to HDRIs that contain clipped high values, "especially the sun". But I also think that replacing the sun area with a 3D light (or as described by Brainspoon) still appears to be the best option so far.



Gerardo
Back to top
 
 
IP Logged
 
tischbein3
Ex Member


Re: Unclipped hdr
Reply #12 - 05/25/12 at 20:41:37
 
Blochi wrote on 05/18/12 at 02:59:46:
I really think it is debatable, and I can see how one can talk about it for a full hour.

Really ? Well how about:

"As for the sun, its simply impractical, period."

Smiley

Ok the longer one of the stuff not discussed here:

1. While driving today, I tried to renember a real life situation where I actually saw the sharp noon-sun-disk in all of is glorry in a low reflective surface, as a non white spot.
Might have encountered some, but then again a regular clipped rendered sun really wouldn't make that big difference.

2. The amount of problems, using nd filters, (either removing it, or having longer exposure times for the rest of the shots) is far greater than the extra centimeter you are gaining in quality.

3. The good old speckle / antialiasing Problems on reflective surfaces.

As for painting out the sun:

I did my tests, and yes,it does improve the quality.
But usually the time is far better invested in other problems.

Back to top
 
 
IP Logged
 
Gerardo
Senior Member
****
Offline



Posts: 74
Re: Unclipped hdr
Reply #13 - 05/31/12 at 04:30:09
 
brainspoon, I mistook the calculation of your HDRI, it's not 19 EVs, it's a bit more of 21 EVs! Smiley   
 
Agree somehow with Chris, the capturing process may be not debatable in a common context, but what may be debatable when integrating pysically-based shaders in real illumination conditions is the reconstruction/reproduction process. Idea in this context is get realistic results in the most direct way as possible, without tweaking and re-tweaking shaders or adjusting a lot of separate buffers in post-processing just to approximate the results we need instead of concentrating in the artistic aspects.
   
In this context, the energy-conserving and reciprocity laws (together with the fresnel effect) make important differences in the way contrast ratios are calculated in real illumination conditions. Even when at a given exposure, a white spot in a reflection may look the same with clipped and unclipped high HDR values, the diffuse response will be different, because there's indeed more light coming from the unclipped values. If we make a simple test with Andreas HDRI (which is about 21 EVs) we can see the difference: 

...

 
I've clipped the original HDR map from 21 EVs to 18.5 EVs. Reflections look the same, but the diffuse component doesn't match. 
 
We can try to compensate by over-exposing the environment map, but notice is not only a diffuse level difference. With exposure compensation, shadows lose their strength, so it's a contrast ratio difference:

...

 
In this case the clipped values are very high, but if they were lower, difference (and compensation) would be more drastic.  In a situation like this,  middle-grays in a reference gray-card won't be really 18%, it would be under-exposed, and we should have to compensate, either by adding a 3D light to reach the real lighting contrast ratio (which should be calculated/captured somehow before), or we would have to tweak our shaders breaking the energy-conserving rule and re-adjusting fresnel coefficients of every surface
 
Still think the 3D light is the best option so far at least in most of renderers, but the importance of reproducing uclipped HDR values somehow is still there when working with physicaly-based shaders.  It seems this has even more sense in the case of PRenderman, where the Physically Plausible Shading system seems to compensate for those bright speckles by firing more rays from areas of the HDRI that really contributes to lightning. Guess this could be similar to the algorithm present in Blochi's Lightbitch.   
 
Btw, last year I shared in a CGTalk thread an experimental technique for avoiding the antialiasing problems on reflective surfaces due to very high contrast ratios.
 


Gerardo

P.D. The server where I stored my images was hacked recently, sorry - here you can find a similar explanation with images  Guess it could be adaptable to other 3D renderers too.
Back to top
 
 
IP Logged
 
brainspoon
Ex Member


Re: Unclipped hdr
Reply #14 - 05/31/12 at 09:19:01
 
Looks great that test!

Right now I stay with capturing standard hdr that range around 18.6 stops during sun light until I can afford better equipment, as I said the results can be improved a lot.

Recently I did a test where I changed the fstop and iso during two brackets and the results came out pretty good.
Photomatix had no problem combining the images and that way I was able to avoid the high capture times.

Now I look for a way to automate this. Would be great if one could program custom capture sequences into the promote control.
Back to top
 
 
IP Logged
 
Pages: 1 2 
Send Topic Print