A CPU path tracer in C++ with BVH acceleration, Monte Carlo sampling, specular reflection, and refraction.
Refractive glass teapot with caustics — 1440p
Specular dragon — 32 samples per pixel
Shiny dragon
- BVH acceleration — spatial median split on alternating axes (composite pattern)
- Möller–Trumbore intersection — fast ray/triangle intersection
- Area lights with stratified sampling — soft shadows via jittered grid supersampling
- Fresnel equations — physically based specular reflection and refraction (Snell's law, total internal reflection)
- Russian Roulette path termination — unbiased stochastic depth cutoff driven by surface reflectance
- Cosine-weighted hemisphere sampling — importance sampling for diffuse indirect illumination
- OpenMP parallelisation — dynamic scheduling over scanlines
Object (abstract)
├── Sphere — analytic ray/sphere intersection
├── Triangle — Möller–Trumbore, Phong normal interpolation
├── Plane — infinite plane
├── AABB — axis-aligned bounding box
└── BVH — recursive bounding volume hierarchy (leaf = triangle mesh)
Light (abstract)
└── AreaLight — stratified sampling over a parallelogram emitter
Camera — generates primary rays through a virtual image plane
Sampler — thread-safe jittered random number generation
WaveFrontParser — loads .obj meshes and builds a BVH
The path tracing kernel (Lo in lambertian_reflection) evaluates the rendering equation recursively, coupling diffuse and specular terms so caustics arise naturally.
Requires a C++11 compiler with OpenMP support (GCC or Clang recommended).
mkdir build && cd build
cmake ..
make -j$(nproc)
./CGProjectThe output is written as a .ppm file in the working directory.
Note: mesh data files (
.obj) are not included in the repository. Place them underdata/before running.
See DEVLOG.md for a chronological record of design decisions, dead ends, and lessons learned during development — from the first ray/sphere intersection in February 2020 through BVH construction, area lights, Russian Roulette, and Fresnel refraction.