Implementation of Monte-Carlo-Path-Tracing

Computer Graphics Course Assignment (Zhejiang University, Fall/Winter Semester 2025)

Features

• Multiple backends: torch, cuda, optix, cpp, open3d_embree
• Single-run rendering and multi-GPU SPP-parallel rendering
• Neural postprocess / denoising pipeline
• Support for real scenes, transparency, cutout materials, and BVH acceleration

Project Layout

• main.py — unified runner
• configs/ — runtime configuration files and backend presets
• modules/, steps/, pipeline/, core/ — renderer implementation
• documents/ — project documents
  • report.md — experiment report
  • environment.md — environment and deployment notes
  • arch.md, algo.md, impl.md, hier.md — architecture / algorithm / implementation docs
  • images/ — figures referenced by the documents

Quick Start

Example single render:

python main.py configs/default.yaml \
  backend=cuda \
  scene.scene_dir=scenes/sponza-converted \
  path_tracer.samples_per_pixel=100 \
  path_tracer.output_dir=outputs/example_run

Example parallel render:

python main.py configs/default.yaml \
  pipeline.mode=parallel \
  pipeline.parallel.gpus=0,1,2,3 \
  scene.scene_dir=scenes/sponza-converted \
  path_tracer.samples_per_pixel=100 \
  path_tracer.output_dir=outputs/parallel_run

Documentation

Please start from:

• documents/report.md
• documents/environment.md
• documents/arch.md
• documents/algo.md
• documents/impl.md
• documents/hier.md