At 06:44 PM 10/30/2005 +0000, you wrote:
> The bmp displacement is 'terraced' because there are only 256 discrete
> heights: the range is 0-255 integer. The hdr version is far superior! Of
> course you have to watch the output range (high dynamics!).
Did you use a Bump object and use the different interpolation methods? This
pixelation of levels shouldn't really be present. The Bump object can
generate a much smoother result at fairly little negative impact to bump
fidelity, and this saves storage capacity and bandwidth when rendering using
lower colour resolution images.
David Coombes
Hi David,
Yes, I used a Bump object and I tried all kinds of smoothing and
interpolations but it was no match for the hdr. And it's strange but file
size of the bmp is larger than the hdr's!
Smoothing sometimes even had an adverse effect, emphasizing the steps...
and if you smooth too much you get a blurred landscape and lose all
crispness. The reason it doesn't work well is probably the large horizontal
distance between the steps. The blurring works horizontally while here it
should work vertically (blurring the slopes). Hard to explain what I mean,
just brainstorming here... but it's obvious the steps are most conspicuous
at weak slopes where they are far apart.
But you have a point: the bmp in my example doesn't utilize the whole 0-255
range, it's too low in contrast; higher quality with bmp is certainly
possible.
WM author Stephen Schmitt mentions this issue, boasting about its 32-bit
precision and I have to confirm it. Now that I think of it, the files I
used were generated from WM's 16-bit Targa output so the hdr's precision
isn't even fully used (indeed I can still see faint steps). Next time FP32 raw!
Obviously, it's unpractical to always use hdr or other float formats for
bumps, but in this kind of situation, where you are in control of the whole
pipeline from start to end, why not.
Thanks for the input!
-Mark H
