I think you don't quite understand the DNKY paper. DNKY's paper only deals with storing the skylight distribution. The methodology assumes you have a method for calculating the skylight everywhere (e.g., some complicated model, or photographs) and suggests using basis functions to store the data efficiently. So,

1. generate a very large data set of intensities from model or photograph
2. calculate the weights. Save the weights and not the original intensities

Then when you need to display the sky,

3. (re)generate the intensities from the weights

You will not get exactly the same sky as if you had lugged around the whole large data set, but if you used enough expansion terms then it should be a good approximation. The DNKY paper discusses steps 2 and 3, not step 1.

This, at least, is my understanding. I have not looked at the paper closely or implemented it.

If you are trying to generate the sky intensities from a model, I suggest considering "A practical analytic model for daylight" (Preetham et al). They very considerately include parameter values and have source code available at their website. The paper has a nice review of the literature too. I have not implemented this either, though.

[edited by - greeneggs on June 19, 2002 12:24:57 PM]

If you guys don''t know about it-
There was a quite popular sky thread a little while back.
http://www.gamedev.net/community/forums/topic.asp?topic_id=86024
It''s not the same technique but it deals with sky anyway and I just wanted to mention it. Yann posts a few screenshots of his implementation with Perlin Noise octaves.

quote:Original post by bishop_pass
Yann L, would you be interested in posting some screenshots of your implemetation? I would be very interested in seeing what your implemetation has produced. There are certain phenomena that I would be curious to see if it reproduces.

Looking at the paper more closely, it is actually all very simple. For skylight, I mean -- I suppose Hoffman & Preetham''s ATI article "Rendering outdoor light scattering in real time" is probably better for the aerial perspective stuff.

I don''t understand one thing, though. There are very simple formulas for computing chromaticities x and y, and luminance Y. But what should be done with these numbers? I guess you can convert to X, Y, Z, and then invert the RGB basis matrix to get RGB basis coordinates. (And then do gamma correction?)

The authors say "the luminance Y and chromaticities x and y can be converted to spectral radiance on the fly using the CIR daylight curve method described in the Appendix." What is going on here? What is the point? What does this give you more than immediately converting to RGB? Sorry, I am not familiar enough with graphics techniques to follow this step.