64k/43k Earth normal map

Tips for creating and manipulating planet textures for Celestia.
Topic author
jim
Posts: 378
Joined: 14.01.2003
With us: 21 years 10 months
Location: Germany

64k/43k Earth normal map

Post #1by jim » 11.10.2003, 21:28

Hi all,

I made some further tests with 64k normal maps. I think the result is not bad. At the moment I have only some parts of Europe maped but the zip size of this tiles is already over 20mb (uncompressed DDS format). So I dont know if I will upload this map. :-(
Note: A unoptimized 64k map in uncompressed DDS format needs 6GByte disk space!

Here are the shots with Waltons 32k surface map and my 8k spec map. :-)
Image
Some river valleys in France

Image
The Alpen with all details.

Bye Jens

wcomer
Posts: 179
Joined: 19.06.2003
With us: 21 years 5 months
Location: New York City

Post #2by wcomer » 12.10.2003, 21:25

Hi Jim,

I've also been building 43k/64k normal maps. I'm building two versions, one at true normal for accuracy and another at 10x true normal to bring out depth. I think the true normal will have a small enough compression size to be reasonably downloaded (at least at least for level[0-5].)

-Walton

Topic author
jim
Posts: 378
Joined: 14.01.2003
With us: 21 years 10 months
Location: Germany

Post #3by jim » 12.10.2003, 23:10

Hi all,

Most parts of North America has now a 64k normal map. At the moment I have round 18% of the world maped but the otimized level6 has even a size of 450mb. I got now the problem that 512mb RAM is sometimes not enougth. 2gb RAM would be really great.

Chris, could it be the Celestia supports once palleted 8 bit DDS format or will this cause some problems with shadder calculation? This format alows 3 times smaller (uncompressed) DDS textures without visible quality losses for normal maps.

Here are the new shots:
Image
I want ever see the Grand Canyon in Celestia ;-)

Image
British Columbia, Canada

Walton, I will see when my map is ready if and how I can made it available.

Bye Jens

bh
Posts: 1547
Joined: 17.12.2002
With us: 21 years 11 months
Location: Oxford, England

Post #4by bh » 12.10.2003, 23:21

Jens...this work looks stunning...I'm still shy of applying any sort of bumpmapping but I'm coming round...great stuff.

Regards...bh.

wcomer
Posts: 179
Joined: 19.06.2003
With us: 21 years 5 months
Location: New York City

Post #5by wcomer » 14.10.2003, 18:24

Hi Jens,

I had success with building true norm VT for level[0-5]. This weigh is at 41MB. Pehaps Mario will be willing to host it.

The level6 of the same had a major problem - contouring. I did not follow Fridger's advice to normalize before scaling. Here is what I had done,

pnmscale -xysize 65536 32768 globe43k.pgm>globe64k.pgm
nm16 [....] globe64k.pgm [...] normal64k.ppm

After a bit of sleuthing I discovered the source of the contouring. Netpbm's upscaling is terrible. It repeats rows for 16-bit greyscale and interpolates columns. Even worse, 24-bit rgb gets repeated rows and columns. Which explains why Don Edwards could see artifacts in my original 43k/64k texturemap. Netpbm has another tool, pamstretch, which will 'interpolate' for upscaling (though not very well.) This tool only works for rgb images. So the best way to go from 43k to 64k with Netpbm is:

nm16 [....] rnglobe43k.pgm [....] normal43k.ppm
pamstretch -xscale=2 -yscale=2 normal43k.ppm>normal86k.ppm
pnmscale -xysize 65536 32768 normal86k.ppm>normal64k.ppm
normrg [....] normal64k.ppm

'normrgb' is a tool I wrote to renormalize the rgb vectors in the scaled normal map, which I would be happy to share. This solution removes the visible artifacts but is far from ideal. I also need to use a similar process to recreate my 43k/64k texturemap.

It seems like Fridger had a similar problem with the interpolation of ImageMagick's 16-bit greyscaling. Looking at your 43k/64k normal map I believe I see the same contouring problem. Check your .dds files and look for horizontal and/or vertical striations/contours.

A nice thing about working with these higher resolutions is that the natural normal map generated from earth's topography is sufficient to generate significant shading effects. Unfortunately the BlueMarble texture already has many natural shadows. So the normal map creates shadows of the same magnitude which can cause unnatural double shadow effects for some lighting conditions. For now, it is probably better to use an exagerated normal map, maybe 10x. In the future I will take a crack at making a tool to automatically detect and remove the shadows using the GLOBE data and maximum likelihood methods.

cheers,
Walton

Topic author
jim
Posts: 378
Joined: 14.01.2003
With us: 21 years 10 months
Location: Germany

Post #6by jim » 14.10.2003, 21:39

Hi Walton,

My way is complete different. First I don't use any tools or normal map filter. I use a combination of Photoshop6 and Photopaint10 and make the hole image processing with the 16bit channel depth. I rescale the 16bit bumpmap to 64k, than use two user defined filters in Photopaint and build the red and green channel of the normal map. After that in Photoshop I optimize/enhance this before convert to 8bit and combine to RGB normal map.

This way is relative expensive and works only up to 16k pieces. Further must I cut this tiles to preven ugly borders.

My source is http://edcdaac.usgs.gov/gtopo30/gtopo30.html . At the moment I must sill some files download. The good news is that I can extract a basic 43k spec map (only the Oceans and some large lakes).

Bye Jens


Return to “Textures”