Project 3: Xuntong Liang

[PacosLelouch]

• Repo Link
• Implemented a CUDA path tracer and completed some performance analysis.
  • Implemented refraction, and non-perfect specular surfaces with Phong material and microfacet material with GGX normal distribution.
  • Used stream compaction thrust::partition after shading to reduce the number of rays for further bounce, and completed performance analysis for this feature.
  • Used sorting by material type before shading, and completed performance analysis for this feature.
  • Implemented a toggleable option to cache the first bounce for re-using, and completed performance analysis for this feature.
  • Implemented importance sampling of scattered ray with different materials.
  • Implemented OBJ mesh loading with tinyObj.
  • Implemented stochastic anti-aliasing.
  • Implemented post processing such as outlining by stencil, and a simple version of ramp shading with a ramp texture.
  • Implemented texture import and mapping with triangle meshes, and HDR image sky sphere.
  • Implemented bounding volume hierarchy as a hierarchical spatial data structure for triangle meshes, and completed performance analysis for this feature.
• Feedback:
  • This framework helps me a lot in how to do bounces separately, and how to use CUDA-OpenGL interop. However, the code structure of this framework seems not so unified. For example, the indents are different in different files. There may be some improvements.
  • Also, I don't know if there are methods to separate the declarations and definitions of CUDA kernels or functions in header files and source files. What I know is that it is difficult to do this with templates, but I'm not sure if so it is with CUDA. A header-only implementation seems hard to maintain.
  • Last but not least, it has a good interop with the guest lecture on Monday! I think there are some idea for further development on the path tracer.