Ray Tracing in One Weekend

A Ray Tracer based on the Ray Tracing in One Weekend Series written in Julia.

This implementation supports GPU parallel processing with the CUDA.jl API. However, not all features from the CPU implementation are supported in the CUDA.jl version.

In order to support several types of objects the GPU compiler requires all types to be plain data types (i.e. collection of primitive types). To achieve that, heavy specialization of methods are performed by the JIT compiler. This implies in a slow compilation, but fast machine code.

Alternatively, multithreading is also supported by using the --threads command line argument when starting Julia. For example:

$ julia --threads 4

Will start Julia with 4 threads.

All necessary packages are included in the Project.toml file. To install the necessary packages from the Julia REPL, enter the Pkg mode (type ]) and run:

pkg> activate <path_to_Project.toml>
pkg> instantiate

Example Usage

To render sample scenes first include the main.jl file and call the render method with one of the available scenes from scenes_list().

For example, the code below looks at the scenes available and render the :cornell_box scene:

julia> include("main.jl")

julia> scenes_list()
KeySet for a Dict{Symbol, Function} with 5 entries. Keys:
  :cornell_box
  :cover1
  :cu_cover2
  :cover2
  :cornell_smoke_box

julia> img = render(:cornell_box);
400/400

After ray tracing the scene, the resulting image is returned as an Array{RGB} which can be used with most image IO packages to store the image. The Images.jl package included in the Project.toml will already have one installed so you can just call the save method to save the image:

julia> save(joinpath("outputs", "cornell_box_1000_samples.png"), img)
┌ Warning: Mapping to the storage type failed; perhaps your data had out-of-range values?
│ Try `map(clamp01nan, img)` to clamp values to a valid range.
└ @ ImageMagick ~/.julia/packages/ImageMagick/b8swT/src/ImageMagick.jl:180
310007

As the Warning explains some values are out-of-range, which is not supported by the ImageMagick.jl backend used to save the image. This happens because the scene is based on the Cornell box code from the Next Week book which has a light with value 15 (see listing 57). To solve this just do as suggested and clamp the image values to [0, 1]:

julia> save(joinpath("outputs", "cornell_box_1000_samples.png"), clamp01.(img))

The saved image is displayed below.

By default the scene uses 1000 samples per pixel (spp) and contains a lot of noise due to rays bouncing out of the scene. We can increase the number of samples per pixel to reduce the noise.

Let's increase the number of samples to 10000 and use the gpu to process the image faster with the spp and gpu keyword arguments.

julia> img = render(:cornell_box; spp=10000, gpu=true);
100.000000%
julia> save(joinpath("outputs", "cornell_box_10000_samples.png"), clamp01.(img))

Using spp=10000 reduces a lot of the noise as shown below.