[Radiance-general] PMAP simulation

[Roland Schregle]

Carsten Bauer wrote:
> Roland Schregle wrote:

>> And Carsten, what *do* you really want? :^)
> 
> Money!  ... and <beep> ! and more money .. and much more <beep> !

Well, I always appreciate frankness. :^)

> back to business: the number of photons is both straightforward and not 
> very helpful at the same time (anyway, it is more intuitive than many 
> other Radiance parameters). But what do e.g. 471100 photons really mean 
> in terms of accuracy or even image appearance?

Nothing -- they're just numbers. That'd be like interpreting the 
semantics of a knitting pattern. At least that's my conclusion after 
dabbling with it for 5 years. :^)

But seriously... there *is* an optimum photon count / bandwidth 
combination, but currently it's still a matter of trial and error. Bias 
compensation is a workaround, albeit a slow one, and not entirely devoid 
of artifacts.

> Additionally, the photon-map currently is not adaptive, distributing and
> gathering are strictly separated. So controlling pmap 'form the output
> end of the line', i.e. the image, is not possible now. 

You're probably suggesting something along the lines of Suykens' density 
control mechanism, which evaluates a local error metric during 
distribution. Of course that requires importance info, which is totally 
lacking in the current pmap implementation, and would require *major* 
revision. With the bare essentials you oh so subtly mentioned above 
(don't forget fame, fortune, learjet, mansion in Malibu, royal suite in 
Dubai's Burj Al Arab, weekends with starlets, stable of Ferraris, 
basement full of Crays, blah blah blah) not in the offing, I'm not 
inclined to invest the significant time and shove this functionality 
into the code. (Like I said, I appreciate frankness.)

Of course, *you're* welcome to try, seeing as you've already delved in 
the code. :^)

-- 
Roland Schregle
PhD candidate, Fraunhofer Institute for Solar Energy Systems
RADIANCE Photon Map page: www.ise.fhg.de/radiance/photon-map

END OF LINE. (MCP)