BBM: A BRDF Benchmark

BBM is a C++20 based BRDF Benchmark that aims to provide:

• Reference implementations of BSDF models

• Support and integrating reading measured BSDFs

• Support BSDF parameter fitting with different fitting methods and metrics

• Easy integration of published fitted BSDF parameters

• Integration in existing Rendering engines

• Bindings to Python

• Support different backbone engines (e.g., enoki, drjit, etc.)

BBM is a templated based library written in C++20 relying on features such as concepts. It has been tested on recent versions of GCC and Clang. BBM uses cmake for makefile generation and sphinx+doxygen for documentation generation.

BBM is still under development. BBM currently (version 0.5.1, August 2023) supports:

• 34 BSDF and variants reference implementations (evaluation and importance sampling)

• support for reading the MIT-MERL measured BSDFs

• python bindings for BSDF models

• a plugin for Mitsuba 1 (RGB and homogeneous BSDFs only)

• 6 different fitting metrics and BSDF fitting support (beta)

• a native backbone implementation that does not rely on external libraries (no autodiff or packet support)

• support for the Enoki backbone

  • fully supported: scalar and packet types

  • experimental: CPU-based autodiff

  • not supported: GPU types

• support for the DrJIT backbone

  • fully supported: scalar and LLVM DrJit backbone for non-differentiable types

  • experimental: CPU and LLVM based autodiff

  • partial: DrJIT python bindings (currently requires manually importing the corresponding drjit.<scalar/llvm> python library)

  • not yet supported

    • DrJIT CUDA backbone

    • packet types

The development of BBM was in part supported by a National Science Foundation grant: CNS-1823154.