amaranthyan / raytracing

Monte Carlo path tracer

RayTracing

My personal project to implement a Monte Carlo path tracer in 1 month following Peter Shirley's ray tracing trilogy Ray Tracing in One Weekend, Ray Tracing: The Next Week and Ray Tracing: The Rest of Your Life.

This project has covered all the base techniques presented in the books. In addition, I added a few advanced rendering features and corrections to the original code, which makes it a (slightly) advanced ray tracing renderer :)

Demo Scene

Cornell box with volumetric clouds and a colored crystal bunny, 1024 ray samples per pixel

Advanced Features

Some avanced rendering features are missing from the trilogy or being presented in an introductory way. In this project, I tried to implement a handful of them while reimplementing the others with more optimized/performant algorithms.

Acceleration Structure

• BVH using strict Surface Area Heuristic(SAH)
  The original SAH implementation yeilds suboptimal BVH structure when there are sticklike objects along the dominant axis, the strict SAH calculates surface areas along all 3 axes to determine the best possible BVH division. Debug marco has also been added to allow visualization of the BVH structure.

BVH of scene from the books, spheres	BVH of custom mesh, triangles

Materials

• Fresnel-Schlick reflectance for metal
• Beer-Lambert absorption for dielectric
  Add ray absorption inside dielectric materials based on the Beer-Lambert law, absorption rate can be tuned for each channel of the RGB to create colored dielectrics.
• Glossy material using Cook-Torrance BRDF model
  Physically based glossy material, each part of the Cook-Torrance BRDF is chosen as follows:
  F - Fresnel-Schlick
  D - GGX Trowbridge-Reitz NDF
  G - Smith GGX Disney
  GGX sampling is used as importance sampling for this material.

Glossy gold bunny, 1024 samples

Cloud Volume

• Worley noise
  Worley noise was implemented in addition to Perlin noise.
• Non-uniform arbitrary cloud volume using 3D noises
  Cloud volume can be created with a mixture of Perlin and Worley noises and the ray sampling in non-uniform medium is done thanks to Woodcock tracking. I also extended the medium boundary code to accept non-convex shapes as boundaries.

Renderer

• optimized random functions
  The original random 3D vector generation in the books adopted a loop and retry strategy, a loop-free method has been put in place to replace it.
• expendable threaded rendering and fixed-size thread buffer memory
  A CPU worker thread pool designed for pixel-level rendering jobs that is scalable to maximize CPU core usage, with fixed yet adjustable memory consumption.
• simple model loading and triangle mesh support
  A simple OBJ model file parser that supports limited OBJ data formats including vertices, normals and texture coordinates.