clipka <ano### [at] anonymousorg> wrote:
> Am 16.03.2017 um 16:06 schrieb muyu:
>
> > I am using the following code:
> >
> > #version 3.7;
> >
> > global_settings {
> >    assumed_gamma 1.0
> >    ambient_light 0.0
> >    max_trace_level 100
> >    adc_bailout 0.0
> >
> >    radiosity {
> >       count 1500
> >       gray_threshold 0.00
> >       error_bound 0.05
> >       brightness 1.0
> >       normal on
> >       recursion_limit 20
> >    }
> > }
> ...
>
> > The recursion_limit has been set as the maximum. But it still a little bit
> > underestimate.
>
> Whether this is to be expected depends on how much energy the canopy
> absorbs per "bounce".
>
> For instance, if 70% of the light is either reflected or transmitted per
> bounce, the error due to reaching the maximum recursion limit should be
> less than 0.1% of the total incoming energy; but if 80% is reflected or
> transmitted, that error already exceeds 1%, and at 90%
> reflection/transmission it reaches 12%.
>
> > 1) Do you have other suggestions to increase the accuracy?
>
> No, not really. The other radiosity settings should just affect
> low-frequency noise-ish artifacts intrinsic to the sampling and caching
> mechanisms.
>
>
> > 2) gray_threshold has to be set as 0.0? I found that this impacts significantly
> > the result.
>
> Yes, a setting of 0 (being the default, and serving as a special value
> indicating that the feature is to be turned off) is the most realistic
> (to the extent that POV-Ray's internal naive RGB colour math is
> realistic, compared with a full-fledged spectral colour model; but you
> seem to be using POV-Ray in a kind of monochromatic mode anyway, so the
> naive math should be fine for you).
>
>
> > 3) brightness seems has no significant on the result. Is it correct to set
> > brightness as 1.0?
>
> If you change brightness significantly, you /should/ see a difference.
> But 1.0 (also the default) is the most realistic setting.


Thanks. 1) Regarding my simulations, around 50% of the light is reflected in
near-infrared. So physically the expected error should be much less than 1%.
This should be neglible?

2) However, I found almost 10%  under-estimates compared with the reference
using Monte Carlo method. What's the possible reasons for this? Should I
increase the brightness or the light intensity?

Post a reply to this message