I’ve done a ray-cast version of the algorithm now and I thought I should breif you on my results.
In most cases the hardware version and the ray-cast version produces very similar results, however, the hardware version is significantly faster, at least in my implementation.
However, in some cases, the ray-cast version produces better results (no huge difference, though).
The hardware version processes data per high-poly vertex while the ray-cast version processes data per pixel (which can be alot at for example 1024x1024…)
I’ve left both version in the tool because I think they both can be useful.
The hardware version can be useful to quickly test your models and see what renderbump-results they produce and the ray-cast version can be used for extremely important models (like a character head).
The tool I’ve been working on is now called version 1.0 (I’ll just have to wait and see if it really can be called that or if it’s full of bugs =)
The raycasting can be greately accelerated by using a hierarchy. The one that produces better results is the kd-tree. In my case a small texture of 512x512 takes only a few seconds, and if you enable antialiasing only a few minutes.
The good thing about raycasting is that it allows you to extract not only the bumps, but any other feature that you want (colors, displacements, etc)
The plugin works nicely on max4, some people have reported problems with other max versions, but I can do very little to solve that. Probably the best thing to do would be to release the code. I don’t have anything against that, but that code is not something that I feel very proud of. I’ve been trying to find some time to clean the most embarrasing parts without suceed. So, if anyone needs the code, just ask me for it privately.
I know the feeling, I could probably spend
the next 6 months improving, adding functionality, etc, on my tool, however,
I got plenty of other things to do, and since
it isn’t a commercial tool, it’s harder
to motivate the time you have to spend.