Limits of DXT encoding for 16bit normal maps
Posted: 13.07.2003, 16:20
With the enormously improved smoothness that one obtains from calculating the texture normals (gradients) from 16bit precision elevation maps, instead of merely 8bit, one seems to be approaching the limits of capabilities of DXT-compression! Same for JPG's but this will surpeise nobody.
Here is a comparison. I displayed a tiny piece of the 8k-sized MOLA normal maps of Mars over a uniformly white main texture. With the default FoV (~30 deg) I display them at only 400 km distance, so a very high magnification is involved. All normal maps are derived from the same 16bit nm16-generated *.PNG master texture!
First the *.PNG original, which is amazing in its smoothness despite the huge magnification:
Next the 24bit-color DXT3 encoded normal map (using nvdxt.exe):
Finally a 85% quality JPG compression
I guess the conclusions are obvious! DXT5 encoding does not give much better results.
Bye Fridger
Here is a comparison. I displayed a tiny piece of the 8k-sized MOLA normal maps of Mars over a uniformly white main texture. With the default FoV (~30 deg) I display them at only 400 km distance, so a very high magnification is involved. All normal maps are derived from the same 16bit nm16-generated *.PNG master texture!
First the *.PNG original, which is amazing in its smoothness despite the huge magnification:
Next the 24bit-color DXT3 encoded normal map (using nvdxt.exe):
Finally a 85% quality JPG compression
I guess the conclusions are obvious! DXT5 encoding does not give much better results.
Bye Fridger