Open Source Renderer Aqsis 1.0 Released

cgTobi writes "I am very pleased to announce that the stable 1.0 release of Aqsis - The Open Source Renderman Renderer, has been released. This release will remain stable in terms of publicly visible interface, no new features, only bug fixes. This will allow users who have been concerned in the past about things changing underneath them to use Aqsis in the confidence that it will not change. We have branched the CVS repository to allow 1.0 to be maintained in terms of bug fixes, while work goes ahead on new exciting features, including performance and memory optimisation, ray tracing/global illumination, and deep shadow maps."

Re:This is cool

[Pseudonym]

Unfortunately, aqsis is not a raytracer, just a scanline renderer, but according to their faq they plan on adding raytracing and global illumination next.

If it helps, Aqsis is about at the stage where Pixar's Photorealistic RenderMan was at about the time of Toy Story. So while it is unfortunate (and we know how we're going to do it), don't think of this as a limitation. :-)

How does this compare to BMRT and blender?

Unlike BMRT, it's available. :-)

Re:This is cool

[jd]

BMRT is dead, although you can download the last compiled binary. It means that if glibc changes too much, you'll need to have an old version installed to run BMRT. It seems unlikely any fresh builds (even with the existing source) are ever going to be made, because the developer got into legal trouble with Pixar after trying to go commercial.

Blender, as far as I know, is just a design tool, it doesn't actually do any rendering, it just allows you to develop models in a range of formats that you can then throw at a renderer.

BMRT was nice in that it did radiosity as well as ray-tracing, which meant that you could get "soft" diffuse reflections, giving you a far more realistic image. I don't know of any other packages which are anything like as good.

An improvement over ray-tracing is "cone-tracing", as that allows you to handle direct reflections that have some element of diffusion. (Which is most of them.) Cone-tracing programs exist, but I don't know of any really good ones.

The "best" (from a graphics point of view) would be to use wave-tracing, as this allows you to handle not only any type of reflection, but also diffraction. As far as I know, there are no wave-tracers out there except MAYBE in the very high-end market. Wave-tracing is very expensive on CPU cycles, which is why people tend to use approximations.

Re:This is cool

[Pseudonym]

An improvement over ray-tracing is "cone-tracing", as that allows you to handle direct reflections that have some element of diffusion. (Which is most of them.) Cone-tracing programs exist, but I don't know of any really good ones.

Cone tracing is a nice idea at first, but it doesn't actually fit well with the demands of a modern high-end renderer:

• The mathematics of tracing objects other than simple ones (e.g. procedurally displaced surfaces) is difficult and usually involves splitting the cone. Once you've split the cone, a few levels, you're effectively ray tracing anyway (more or less).
• It more or less solves the problem of integrating in four dimensions (two screen-space dimensions and two lens-space dimensions), but does nothing to help the problem of the other two interesting dimensions that you tend to sample stochastically (i.e. time, for implementing motion blur, and continuous level of detail). The problem is even worse in a full-spectrum renderer, where wavelength is also sampled stochastically.

The "best" (from a graphics point of view) would be to use wave-tracing, as this allows you to handle not only any type of reflection, but also diffraction.

Not polarisation, though. Well, not yet, anyway.