raytracer

A toy raytracer I wrote using Go, in a fit of pandemic-induced boredom. I've never formally studied computer graphics so it was a neat introduction to many concepts as well as a good refresher on basic geometry and vector math.

Geometric objects

The raytracer supports the following types of scene objects:

• Planes
• Spheres
• Discs
• Boxes (rectangular prisms)

Lighting and shading

The raytracer simulates two different kinds of light sources:

• Point lights, which have a defined location and cast light omnidirectionally, and
• Distant lights, which illuminate surfaces from a fixed direction.

Soft shadows are simulated by randomly varying the location/direction of a light across many samples.

Surfaces have a diffuse component, a refractive component, and a reflective component. Only two kinds of diffuse textures are currently supported: solid colors, and an alternating "checkerboard" pattern of two colors.

Other rendering features

Anti-aliasing

To prevent jagged lines from appearing where the scene has edges, the raytracer supports supersampling, in which multiple rays are cast for one pixel and the results averaged together.

Depth of field

The raytracer can produce a "depth of field" effect (blurred foreground/background) by simulating a camera lens having a non-zero-radius aperture and a finite focal distance; multiple rays are cast back from the focal plane through random points in the aperture, and the results are averaged together.

Parallel processing

The rendering algorithm divides the image into multiple work units and distributes them across a pool of worker goroutines, using channels for coordination.

Animation

The binary takes a -frame parameter which can be used in the scene setup code to vary any parameter over time. After rendering each frame into a separate PNG, you can use a tool like FFmpeg to combine the separate frames into a video. Here's an example animation of the focal distance changing from the frontmost sphere to the rearmost sphere and back:

Missing features

Some of the obvious features this raytracer doesn't support are:

• Reflections of lights off of reflective surfaces
• Refraction of shadow rays (rays cast from a shaded point to a light source)
• Performance optimizations to not supersample pixels that don't need it (currently, rendering a 4K version of the scene above takes several hours on a quad-core PC)

Acknowledgements

I found the articles at scratchapixel.com really helpful for understanding many of the concepts underlying a raytracer.