Mitsuba 3 is easy to install and use from Python, which simplifies many things. Materials, textures, and even full rendering algorithms can be developed in Python, which the system JIT-compiles into efficient megakernels for the GPU (via OptiX) or CPU (via LLVM). 2/10
Most of the system can be customized/replaced to facilitate experimentation. The target audience are researchers in graphics, vision, and fields investigating inverse problems involving light. 3/10
Mitsuba 3 also implements recent differentiable rendering algorithms developed by our group (e.g.
@DelioVicini's path replay backpropagation). The most tangible improvement for users is that Mitsuba 3 can differentiate complex scenes without running out of GPU memory 😅. 4/10
Finally, we started a video series to explain Mitsuba (and soon, Dr.Jit) in an accessible way. There are only a few videos so far, but we'll be making a bunch more. You'll also get to meet the team members working on the project. Intro from me (#1): tinyurl.com/mitsuba-tut 6/10
These cover basic steps, but we'll be working our way up from there—stay tuned! To get a taste for more, you can find complete Juypter notebooks covering a variety of applications, how-to guides, and reference documentation at mitsuba.readthedocs.io. 9/10
The docs already include tutorials for caustic design, inverse volume rendering using delta tracking and "NeRF" grid-based radiance fields. It feels good to release this project, which has been in the works for quite a while.. We're excited to see what you will do with it! 10/10
