Celestia Forums

Some of the normalmaps seems to have been generated from existing bumpmaps. Is it so? And if yes, how many of them is generated that way?

TriKri wrote:Some of the normalmaps seems to have been generated from existing bumpmaps. Is it so? And if yes, how many of them is generated that way?

Normalmaps are generated from elevation maps that may also serve for bumpmapping. What did you think how they are made? We have special software to convert 16bit integer elevation maps to highest quality normal maps.

Bye Fridger

Well, I thought maybe you had a png image, or something similar, but 16-bit values seems accurate enough. I read something about someone had generated a normalmap, And I thought, was it from an bitmap or what?

A normalmap can be made from almost any image format, JPG, PNG, BMP, TIFF, TGA, PSD, ect...
As long as the image has has the needed hight values and you have the tools to make them with.

Don. Edwards

Certainly, it's true what Don writes.

But the main challenge is to generate normalmaps that are VERY smooth AND have virtually NO NOISE.

In order to achieve this, it is crucial NOT to use height maps that had been converted to 8bit (from their published 16bit accuracy) before being encoded as JPG, PNG etc. Some of the popular formats are simply not able to carry 3x16bit RGB information!

Obviously, only 256 height values (8bit) for a whole planet are not sufficient, while a range from -32768 m to 32768m (signed 16 bit) is!

All /professional/ height map data for astronomical bodies are published as 16bit graymaps.

Bye Fridger

Ok, I see. But what format has 16-bit accuracy? I supose you could use bmp or png if you only mapped each color to a value, then you would get 24-bit height maps, is that technique used?

Then I supose jpg could contain enough information if you put down the "compression rate". Then it can contain more information than a bmp image, it is just that when it comes to computer screens, they can't show more than 256^3 colors anyway. True? Is jpg used in that way?

TriKri wrote:Ok, I see. But what format has 16-bit accuracy? I supose you could use bmp or png if you only mapped each color to a value, then you would get 24-bit height maps, is that technique used?

I don't understand your logics why this should be useful? The input height maps are first of all experimental data. The precision that is used is determined by the possible precision of the altitude measurements. 16 bit signed integers are just adequate. Our recently finished nmtools distribution does all required operations at the 16 bit level and only converts to 3x8 bit normalmaps in the last step. Hence no accuracy is vasted in rescaling, forming gradient (i.e. normal) vectors etc.

I wonder why you keep returning to particular formats? We use the natural format (16bit int) that is generally appropriate for the input data. The output normalmap is 3x8 bit RGB anyway and may be trivially converted into any convenient standard format by the usual tools. Celestia requires 3x8 bit for normalmaps.

Then I supose jpg could contain enough information if you put down the "compression rate". Then it can contain more information than a bmp image, it is just that when it comes to computer screens, they can't show more than 256^3 colors anyway. True? Is jpg used in that way?

What do you want to say or ask? Normally we use lossless PNG format for normalmap VT tiles. They give the best quality. Recently there is some new progress concerning a particular DXT5 compression scheme for normal maps that I was discussing in my respective Developer Talk thread. That is now supported by Celestia in CVS.

Bye Fridger