Celestia Forums

Hi all,

since some time I am contemplating about how to implement better
i.e. more instructive images of our star, the sun in Celestia!
There are these great images from SOHO, at 6 different wavelengths at a given time of observation, altogether a
wealth of amazing information...

Have a look how great they are (just 3 wave lengths):



Recently, I had this idea of exploiting mipmaps for that
purpose! Mipmaps are always generated and stored along with the main
texture in DXT (dds) formatted textures! What does that imply?

Take a 4k solar texture, for example. Along with the main 4k texture,
the dds texture would then store a sequence of 12 mipmaps, each being
a factor 2 smaller in (linear) size than the previous one. The point is that
depending on the distance of observation of that texture, lower and
lower resolution mipmaps are smoothly blended in during the
display. After all, at large distance from the sun, we cannot
distinguish all those fine details anymore!! So we save resources this
way.

My idea is based on the fact that e.g. with the NVIDIA command line tools
one may assemble custom made mipmap images into one dds
texture!

This could be exploited as follows:

Take some usual low-res standard solar texture, just with a few sun spots for
the large distance mipmaps, but incorporate mipmaps from SOHO for
short-distance hi-res display! The blending from low-res to hi-res
SOHO stuff would then be done automatically.

Of course I am always getting back conceptionally to my other proposal
of incorporating Celestia virtual "instruments" through a set of filters for
different light wavelengths. The SOHO data would be great for this
since with a virtual "filter wheel" one eventually might tune through
the various data at different wave lengths (taken at the same time!)...

Bye Fridger

Hello again Fridger,
Do I unserstand correctly that mipmaps are sort of like whole-boldy virtual takes. ie, they don't have higher resolution of one area, but of the whole body that are blended in? If I understand, this sounds like a great idea. I wonder about people with slower computers though, how would this affect performance vs. static textures or virtual texture tiles?

alphap1us wrote:Hello again Fridger,
Do I unserstand correctly that mipmaps are sort of like whole-boldy virtual takes. ie, they don't have higher resolution of one area, but of the whole body that are blended in? If I understand, this sounds like a great idea. I wonder about people with slower computers though, how would this affect performance vs. static textures or virtual texture tiles?

Joe,

mipmaps are just part of the default DXT format setup. (Of course you may always suppress the building of mipmaps with a switch). Any dds converter generates these mipmap images automatically by rescaling the original texture by powers of 2 successively. So normally, the mipmaps are just lower resolution clones of the original main texture.

But the individual mipmaps may be modified by hand, and it is here that my above idea comes in.

I was also wondering, whether mipmaps might be profitably used to model different layers of atmospheres when descending to the surface of a planet. But since Chris will recode the atmosphere stuff completely, let's wait and see, first...

Bye Fridger

Hm... aren't mipmaps used also for near horizont rendering on curved surfaces? How is this handled in Celestia? Or is it an intrinsic functionality of the OGL library. Could create funny effects if so. Those very eyecatching flares could be a problem, too.

But apart from that, I'd realy like to see a more flexible handling for alternate surfaces (also for stars). And I also strongly espouse that virtual instrument concept. As you perhaps know, I tried to support the celestian community with such sun impressions in an addon - but that was less more than popping up that static and some moving pics aside the sun - no real solution.

maxim