Hey all :-)
What parameters do I need to change if I would analyse a small room in a fairly large complex scene and still get a precise result?
Is it possible to change other parameters than -ab -ad -as -ar -aa?
Is it possible to make it change dynamic depending on the scene size?
The five parameters you list are going to be the most important to modify based on your scene. They control the sampling rate of the environment and ambient interpolation between samples. The values they should be are dependent on your simulation scene!
Things to consider:
- ad, ambient divisions
This parameter controls how many rays are traced for each sensor-point/pixel and reflection. Increasing it means your environment will be sampled more rigorously. The exact value this should be will depend on the sensor distance from the important objects that will need to be hit. For example, if you have a very large space with venetian blinds, you need enough ambient divisions to pass through the blinds and also hit them.
- as, ambient supersamples
When a significant difference is found between neighboring samples, this controls the amount of extra rays case to resolve the difference.
- aa, ambient accuracy
This is set in terms of percent error due to ambient interpolation. A value of 0.1 means, roughly, that 10% difference between neighboring ambient divisions is ok without additional sampling.
- ar, ambient resolution
This is probably the trickiest value to set, and it controls the density of ambient values used in interpolation. If you set it to 0, the re is no interpolation.Greg Ward says that "Error will start to increase on surfaces spaced closer than the scene size divided by the ambient resolution." The scene size defined by Radiance is, I believe, the maximum length of the scene bounding box. So if you have a 1000 unit long scene with lots of 1 unit long surfaces, you might want to set the ar equal to 1000 such that no interpolation will occur between your surfaces.
Thank you so much, I forgot to thank when i read it first time :-)
I have an additional question :-) (I can't get the right settings)
If I e.g. have a landscape 350000 x 350000 units and I like to analyze the daylight factor in five rooms (e.g. 6000 x 6000 units). The rooms are in narrow streets ... What would my settings be?
-ab? -ad? -as? -ar? -aa? (time pr. rendering ~30 min)
The results should be presice enough to give the same result if compare to RadianceIES.. (+/- 0.3 DF)
I have attached two example files..
I think it is hard to get a fool-proof recommendation from someone else unless they spend quite some time with your scene. As for what the "right" settings are.. you should try to tweak (and understand them in some way; few ever totally "get" them, including yours truly) until the results do not radically change anymore. Say, you start with 5 bounces, test 7 on a part of the scene with a few sensors and determine that there is no difference- well, then 5 are enough. I don't get what units you are using- I think metres are customary, so you'd use the formula outlined here ( http://daysim.ning.com/page/keyword-ambient-resolution ) to determine -ar. (which in turn also relies on -aa).
All of this takes patience and a lot of experimentation, I am afraid.
Thanks for the answer.
I'll try to iterate me to the right settings.
I just thought there was some rules of thumbs. Maybe a spreadsheet or something like that..
Of course there are things you can calibrate between Radiance and IES. The number of ambient bounces in Radiance should probably be one more than in IES (this is because of a strange way the ambient contribution is handled). Its pretty safe to fix aa at 0.1. Increasing Ad will increase the accuracy of course, -- you can quickly determine what it should be by gradually increasing it with a low bounce (ab =1) calculation until you don't see any improvement. Ar is the easiest to set in that you can simply divide your scene size (350000) by the size of the smallest object you need to account for. If its a 5 unit mullion, that gives you an ar of 70000, but its probably safe to start at a value like 5000 and work your way up until you don't find any simulation differences.
Sorry this isn't more clear-cut.
Thanks again Alstan!
it's a huge help! :-)