Comparison of 32k vs 43k Bluemarble

[wcomer]

Hi Folks,

I conducted a little experiment and built a level6 addition to the 32k Bluemarble texture. The reduction of the 21600x21600 .tif's into 16kx16k textures caused a loss of ~42%=(1-(16384/21600)^2) of the information contained in the original BlueMarble data. I wanted to know how much of a visual difference this made. So I went the other direction and blew the 21600x21600 up into 32kx32k textures and built a level6 set of 512x512 VT tiles. These contain ~74%=(21600/16384)^2 more information than the level5 tiles. Here is a comparison:

Grand Canyon in 32k


Grand Canyon in 43k/64k


For me the difference is very noticable when switching between the 32k and 43k/64k altsurfaces in many parts of the world. However for some of the world the change is not so remarkable (like much of france, england and germany because the terraign is rather featureless... no offense to those who live there. .) The difference is enough so that I will be making 43k/64k specular and normal maps.

Unfortunately the level6 zip (after removing water tiles and compressing higher latitude widths) is ~200MB. So I doubt this can be hosted anytime soon. For those with the ability you should consider building the 43k/64k set.

Likewise, I hope that Don Edwards is making a base 43k texture so that at some point in the future it can be blown up to a 43k/64k VT and hosted. I would much rather be using his work than the BlueMarble base.

cheers,
Walton

[Don. Edwards]

One problem with stretching a texture to the next size up is it adds some distortion to the texture. Pixels are getting stretched that are not meant to be stretched and this means this is really a false texture and technically inaccurate. That is why most of the texture creators never stretch a texture up in size. When push comes to shove it has always been better to go down. If Celestia had the ability to use odd sized textures this wouldn't be an issue. But I can plainly see the distortions created by increasing the size of the texture.
Also I am not working with my texture at the higher 21k level. I simply do not have the system resources to do this. I am pushing my systems limits now. I can't even open or work on a 32k texture in Photoshop in true 32k x 16k form. I also can’t do this in GIMP. With amount of memory I have I am limited to a 30k x 15k texture. It just will not give me the extras headroom needed. Besides I don't know about you but it taking 5-7 minutes just to open a 16k x 16k half texture is daunting and then depending on how much work I have done on the texture it can take as long as 10 minutes just to save the work I have done. You can see why it is taking me so long. For me to be able to crank out these textures any faster I would need 3Gigs of RAM. That is double what I have now an right now with the financial crisis I am in, that is just out of the question. I can't even be sure if I will have internet access next month at this point. So I will keep striving to create a very nice 32k texture for level 5 virtual textures and resize it for the lower resolutions. I will never stretch a texture to make it work except when it is my oceans texture of course.

[wcomer]

Hi Don,

It was mostly wishful thinking on my part that you might have been working at the 43k resolution. Obviously reality does not permit such fantasies.


The purpose of the level6 tile set is not to provide an interpolation for 64k, but rather to include all the information contained within the 43k so as to decrease the minimum viewing altitude before 'inaccuracies' are introduced. I believe that the level tiles I created adequately meets this goal.

While I'm not an imaging expert, I have done just enough signal processing work to be dangerous, but not enough to be 100% certain of what I'm about to say. My understanding is that stretching the texture to 64k and then viewing the texture at a altitude such that the pixels are one-to-one with the screen at a 43k resolution (roughly the second image I have shown) should produce an image that is for all intents and purposes imperceptible from the original 43k image. If I upscale 43k to 64k and then downscale to 43k and diff, the mean square error should be on the order of the magnitude of the least significant bit (assuming that decent algorithms have been employed.) This is something that most people cannot detect with their eyes. Now with respect to viewing at progressively lower altitudes this will, as you have said, introduce 'inaccuracies.' However, these same 'inaccuracies' would also exist if one uses the 32k texture, moreover the 'inaccuracies' in the latter case will be far worse.

I can certainly understand your resistance to and distaste for the notion of upscaling as it does introduce the risk that user will believe that what they see at the 64k resoltution is reality rather than a lesser interpolation of the same.


Walton

[HankR]

Just FYI, the sample image is not the Grand Canyon. It's the Glen Canyon (Lake Powell)/Canyonlands area in Utah, upstream from the Grand Canyon.

- Hank