Project 3: Jefferson Koumba Moussadji Lu

[JeffersonKoumbaMoussadjiLu]

Repo Link

Link To Repository

Core Features Completed

• Shading kernel with BSDF evaluation (diffuse, perfect specular surfaces)
• Stream compacted path termination
• Sorting by material type
• Stochastic sampled antialiasing

Extended Features Completed

• Visual Improvements

  • Refraction with Schlick’s Fresnel approximation
  • Depth of Field via thin-lens camera model
  • Direct Lighting using next-event estimation + shadow ray occlusion
  • Post-processing with ACES filmic tonemapping
  • Better random number sequences for Monte Carlo

• Mesh Improvements

  • ....

• Performance Improvements

  • Russian Roulette for probabilistic path termination
  • Efficient parallel stream compaction (GPU prefix-sum)
  • Radix sort for material sorting (custom CUDA implementation)
  • Shared memory optimization in scan and scatter steps

• Other Improvements

  • Camera Motion Blur (simulated shutter exposure for moving camera)
  • Scene modularity and composition via JSON configuration files
  • Multiple original scenes created (e.g. Solar System, Voxel Castle, Tower, etc.)
  • A somewhat working Restartable path tracing

Other Features and Details

• Used custom stream compaction kernels from Project 2 (naive.cu, efficient.cu, thrust.cu) directly integrated into the renderer.
• Modified CMakeLists.txt to enable stream_compaction module and link correctly.
• Full replacement of Thrust for prefix scan, scatter, and material-based sort.
• Implemented random engine using pixel-index-dependent hashing + low-discrepancy LDS sampling.
• Carefully tuned ray epsilon offsets and importance sampling strategies.
• Extensive JSON scene definitions used for procedural composition and testing.
• All shaders, camera behaviors, tone mapping and integrator toggles are runtime-controllable.

README Completion Checklist

• Cover image in readme does not use Cornell Box
• Descriptions, screenshots, debug images, side-by-side comparisons of features implemented
• Analysis
• Scenes and meshes included or linked
• Third-party library changes or compilation changes documented
• Bloopers (optional)

Late Days Used

• 3 late days used for code
• Code submitted 10/05/2025)
• Read.me submitted (10/07/2025)

Project feedback

Loved the balance of visual fidelity and CUDA-level optimization in this project. The structure encouraged both graphics innovation and GPU best practices.

Note:
Sorry, I just noticed that my pull request description was incomplete, and I apologize for that.
I hope this didn't cause any issues.
Best,
Jefferson