Fridger,

On Sun, May 31, 2009 at 9:40 AM, Fridger Schrempp <fridger.schrempp@desy.de> wrote:
Chris,

Chris Laurel wrote:

>     And so on...
>
>
> I tire of your persistent subtly insulting tone.


Sorry, this remark of mine was not at all meant to be insulting. I rather felt
that a EVEN more detailed discussion of this earth texture (of which you had
left out ALL technical background information including its author) would lead
too far, since you indicated anyway that it was NOT for 1.6.0. My remark "And
so on" was just to cut my own writing short.

OK, understood...

As to the issue of DXT5 vs DXT3 I have several detailed independent
investigations besides my own. I have tested the corresponding improvements in
alpha channel compression for our typical demands in MANY cases with
nvimgdiff. In all cases there was a clear advantage for DXT5 over DXT3.

I trust your results, but I'm still surprised that DXT5 gives better results for specular masks than DXT's explicit alpha. I tried nvimgdiff on a DXT3 and DXT5 version of the 4k Earth texture:

DXT3:
  Mean absolute error: 0.276043
  Max absolute error: 8.000000
  Root mean squared error: 1.200278
  Peak signal to noise ratio in dB: 46.545162

DXT5:
Alpha:
  Mean absolute error: 0.210957
  Max absolute error: 23.000000
  Root mean squared error: 1.206244
  Peak signal to noise ratio in dB: 46.502098

At this resolution, the RMS error is actually slightly (and probably negligibly) less for the DXT5 version. The max absolute error in the DXT5 version is not surprisingly worse than for the DXT3 version, where the 4-bit quantized alpha will naturally limit the absolute error. As to which is nicer visually, I don't know, but I'd guess that the larger absolute errors have a bigger impact than small difference in RMS error.


Based on the fact that Earth is in a way of special importance, I at least
would have appreciated to receive a more concise "CV" of your earth.dds
besides that "naked" link. The image data about earth are so good that any
more serious discussion requires much more information to start.

The Earth image that I provided was not intended to be the beginning of a serious discussion about creating a new Earth texture. I probably shouldn't have linked to it at all. The important statement was the one about waiting until 1.6.1 to create a new Earth texture, a matter on which we both agree.
 
> I do like the appearance of your Earth better than one in the current version of Celestia.
> What about switching to this new set of Mie parameters and Earth texture simultaneously in
> 1.6.1?

That looks like a sensible option...

Fridger

Good. I'm hopeful that RC3 will become 1.6.0 final then. I'll wait until Pat's done with the changes to the GTK version before announcing the official release.

--Chris