functional-pathtracers

Code by Joyce Lee

OCaml and Haskell implementations of a basic Monte Carlo pathtracer

Inspired by CIS 460, 120, and 194 at UPenn

What is it?

Path tracing is a computer graphics Monte Carlo method of rendering images of three-dimensional scenes such that the global illumination is faithful to reality. Fundamentally, the algorithm is integrating over all the illuminance arriving to a single point on the surface of an object.

The above definition is courtesy of Wikipedia, and does a very good job of summing up what it is. It implies that the algorithm terminates when every pixel value converges, but in practice this is intractable. Instead, a maximum recursion depth is specified instead.

Why functional?

The concept of pathtracing is mostly recursive, and simply iterated over a two dimensional array of pixels, which lends itself nicely to functional programs. They usually are not written in languages such as OCaml or Haskell because these run many times slower than languages like C or C++, even with compiler optimizations. However, the functional implementation is arguably more elegant and easier to understand.

How To Use

Both the Haskell and OCaml versions provide a hardcoded scene which will be rendered if no arguments are given. The two of them have slightly different command line arguments though.

• OCaml: compile with ocamlopt pathtracer.ml and run with

  1. ./a.out width height samples
  2. ./a.out width height output-file
  3. ./a.out width height output-file config-file

  The default output_file is ocaml.ppm.

• Haskell: compile with ghc -O2 pathtracer.hs and run with

  1. ./pathtracer
  2. ./pathtracer width height samples
  3. ./pathtracer width height samples config-file

  The default width, height, and samples are 100, 100, and 60 respectively. The output is always written to stdout.

To make your own configuration, each line represents a sphere. The lines should be formatted as such:

radius centerX centerY centerZ emittedR emittedG emittedB colorR colorG colorB [Diff/Spec/Refr]

Check it out!

Take a look at the samples:

1. animation.gif (OCaml) post-processed in C to denoise with a small Gaussian kernel and compress
2. example.png (OCaml) and example-hs.png (Haskell) note the OCaml division by zero artifact
3. green.gif (OCaml)
4. pathtr.png (Haskell)
5. specular.png (OCaml)

To Do

• Artifacts in Haskell implementation due to non-uniform sampling
• Make it faster?