I have recently come to the conclusion that the artefacts visible in normal maps, specifically over desert regions, are not actually artefacts introduced from resizing at all!(as i think fridger may have suggested - i could be wrong here)
With the added increase in normal map quality while using the uncompressed dds format these 'artefacts' actually take on some shape. See the pictures below...
In the above picture, the top ~3/4 of the picture has a 32k uncompressed normal map and the bottom ~1/4 has a 16k dxt3 normal map. It's as clear as day that the artefacts in the 16k compressed map resolve themselves into sand dunes in the uncompressed version. Most of the GLOBE DEM data around the sahara region is of low resolution, as can be seen from the dunes visible on a large scale in the centre portion of the picture(low resolution) and on a small scale in the top portion(high resolution).
A similiar situation exists for the himilaya region were most of the height data of the desert seems to be high res...
These 'non artefacts' are also most visible when using uncompressed normal maps for mars(where the whole planet is now covered in them) . In this case though using dxt3 virtually eliminates them(they are quite ugly)...
Uncompressed u888
Compressed dxt3
Perhaps the limitations of the compressed dxt format are useful after all.
Normal Map Artefacts
Normal Map Artefacts
Slan
- t00fri
- Developer
- Posts: 8772
- Joined: 29.03.2002
- Age: 22
- With us: 22 years 7 months
- Location: Hamburg, Germany
Slan,
quite a while ago, I have done vast amounts of studies of 16bit/channel, 16k normal maps of Mars. So what you write above is mostly known to me. I have never claimed that the artifacts are introduced generically from resizing. It's more subtle: Since we have to resize to powers of two somewhere, the issue was where do to the resizing for best quality! What I pointed out was specially related to using the 16bit/channel raw elevation data. You will get horrible artifacts if you resize the raw data before converting them to a normal map using Chris' nm16 tool! These artefacts are of a much worse kind than what you illustrated above!
I have not studied uncompressed u888 format at the time since Celestia did not yet support it. Instead I used PNG which also does not introduce bad extra 'features'. My results match your u888 ones in quality.
Bye Fridger
quite a while ago, I have done vast amounts of studies of 16bit/channel, 16k normal maps of Mars. So what you write above is mostly known to me. I have never claimed that the artifacts are introduced generically from resizing. It's more subtle: Since we have to resize to powers of two somewhere, the issue was where do to the resizing for best quality! What I pointed out was specially related to using the 16bit/channel raw elevation data. You will get horrible artifacts if you resize the raw data before converting them to a normal map using Chris' nm16 tool! These artefacts are of a much worse kind than what you illustrated above!
I have not studied uncompressed u888 format at the time since Celestia did not yet support it. Instead I used PNG which also does not introduce bad extra 'features'. My results match your u888 ones in quality.
Bye Fridger
-
- Posts: 1034
- Joined: 16.12.2002
- With us: 21 years 11 months
- Location: People's Republic Of Cork, Ireland
-
- Posts: 1034
- Joined: 16.12.2002
- With us: 21 years 11 months
- Location: People's Republic Of Cork, Ireland