John VanSickle nous apporta ses lumieres en ce 2007/09/21 18:23:
> Kenneth wrote:
>> John VanSickle <evi### [at] hotmailcom> wrote:
>>
>>
>>> http://graphics.pixar.com has a paper on combining focal blur,
>>> anti-aliasing, motion blur, specular highlighting, area lights, and
>>> blurred transparency into one feature.  To improve the quality of any of
>>> these, one merely increases the anti-aliasing level.
>>
>>
>> Thanks for reminding me of this link; I had seen it recently in 
>> another news
>> thread comment (probably from you!) but hadn't looked at it carefully.
>>
>>> The way POV-Ray handles this now is rather cumbersome.  If I set 4x4
>>> anti-aliasing, and uses a 4x4 area light, the area light is fully
>>> sampled for each sample of the spatial anti-aliasing, and if I throw in
>>> 10x oversampling for motion blur, then the scene takes 2560 times as
>>> long as a scene with no area lights, anti-aliasing, or motion blur.
>>
>>
>> OW!  That hurts.
>>
>> Thanks for the reply; your explanation of the PIXAR method is 
>> fascinating.
>> Think a similar methodolgy could make it into POV?
> 
> The caveat is that in the Pixar paper, for each ray traced from the 
> viewpoint, only one shadow ray is shot per light source, and only one 
> reflective ray and one transparent ray gets shot.
> 
> The shadow ray is shot to some random point on the surface of the 
> light_source; this neatly simulates an area light of any size or shape, 
> and even one with different colors across its face.
> 
> The reflected ray can be given the strict reflection used in POV-Ray for 
> a smooth reflection, or its direction can be jittered to simulate a 
> blurred reflection.
> 
> Transparency can be calculated directly from the laws of refraction, or 
> the refracted ray can be jittered in order to simulate a translucent 
> interior.
> 
> Motion blur is set by selecting a random time within the time slice that 
> represents the duration of the frame, and positioning everything 
> according to that precise time for that ray shot.  There are tricks to 
> prevent a complete re-parsing of the entire shot for this, and there are 
> shortcuts that trade unneeded precision for speed.
> 
> Focal blur is done by selecting a focal point and then moving the origin 
> of the ray by a random amount for each camera ray.
> 
> With the AA level set to 1x1, the results will be a very grainy image 
> where these features are in force, but as the AA level climbs, the grain 
> gets averaged out.  There are sampling methods that can reduce the 
> graininess as well.  The paper reports that they seldom need more than 
> 64 samples per screen pixel.
> 
> The present POV-Ray method can involve an insane amount of rays shot, 
> and the payoff is not necessarily cost-effective, because the 10x10 area 
> light, necessary to make the shadows smooth, and the 4x4 spatial 
> anti-aliasing necessary to remove the jaggies, multiply together to 
> create 1600 shadow rays per pixel, which is probably overkill.
> 
> Regards,
> John
That's why you have adaptive area_light and aa sampling (methode 2 or +am2). 
It's also why the camera with focal blur always use an adaptive sampling.
This area_light will render faster than a 10x10 without adaptive:
light_source{10 rgb 1 area_light 2*z,2*x 17,17 adaptive 1 circular orient jitter}
It also gives you smoother shadows. With adaptive, 65x65 is still prety fast, 
even 129x129 is fast...

-- 
Alain
-------------------------------------------------
You know you've been raytracing too long when you even think about using Povray 
for writing letters.
Sven Rudolph (Germany)

Post a reply to this message