CPSC 453 HW4

Benjamin Thomas, 10125343, T04

Building/Running

There are two ways to build and run the ray tracer depending on whether you want to use it with the --yours and --default arguments or with much more flexibility. To build and run the simplified version:

1. cd into the wrapper directory
2. Run make
3. Run ./rayTracer --default or ./rayTracer --yours to generate images

To build without the wrapper so that more complicated command-line arguments can be used:

1. Create and cd into a new build directory
2. Run cmake -DCMAKE_CXX_FLAGS="-O3 -DNDEBUG" [path to project root]

• While it is possible to build and use the raytracer with optimizations disabled, the performance penalty from doing so is quite extreme; thus, use of compiler optimizations is highly recommended

3. Run make
4. Run ./hw4 [options] <scene file> to generate images (for more information about available options, run ./hw4 -h)

Features

The raytracer uses a modified version of the scene file format used in HW3 to load its scenes. Thus, it is capable of handling any arbitrary scene made up of triangulated meshes and sphere. In terms of scenes and rendering, the ray tracer supports:

• Any number of perspective cameras with per-camera position, orientation, and horizontal field of view
• Grid-based supersampling for anti-aliasing
• Controllable maximum recursion depth
• Controllable recursive ray bias
• Spheres and OBJ models as objects
• Phong illumination and shading with any number of point lights
  • Controllable light attenuation coefficients
• Diffuse textures read from image files
• Shadows calculated using shadow rays
  • Partially occlusive surfaces (in terms of lighting and shadows)
• Fully and partially reflective surfaces using perfect specular reflection
• Fully and partially refractive surfaces (so long as they are well-nested)
  • Simulation of total internal reflection
  • Note: Does not simulate Fresnel reflection

Internally, the raytracer supports several techniques for speeding up the rendering process:

• Automatric construction and use of bounding volume hierarchies to speed up ray intersection finding
  • One-per-scene object BVH
  • One-per-model triangle BVH
• Parallelism through splitting an image into 8x8 pixel "patches"

The raytracer also prints a small preview image to the console (so long as your terminal emulator supports True Color) and provides a progress indicator during rendering to allow the user to knowwhat percentage of patches have been rendered.

Test Scenes

Three different test scenes are included with the ray tracer that demonstrate its functionality. The first is a basic Cornell box test scene with a reflective sphere and translucent cube. The second is a fully-modelled chess set, as used in HW3, with reflective and translucent pieces. The third is a basic demonstration of a sphere used as a refractive lens in front of a knight.

Cornell Box

The Cornell box test scene demonstrates basic functionality. The differently-coloured walls make it easy to see how the refractive cube is bending light, with the right side of the cube showing total internal reflection. It also demonstrates that sphere normals are calculated correctly, as shown by the reflective sphere.

The shadows below the cube and sphere also demonstrate how shadows are calculated. The presence of ambient light, which is unaffected by occlusion, is clearly visible underneath the sphere. The ability to have objects which only have partial shadows is demonstrated by the light shadow seen under the cube.

A 512-by-512 pixel render of the Cornell box test scene can be found in images/default.ppm. The scene itself is found in scenes/cornell.scn.

Fully-Modelled Chess Set

The chess set scene demonstrates a lot of the more advanced functionality of the ray tracer. All objects in this scene are made up of triangles loaded from OBJ files. The table, chairs, and chess board use diffuse textures loaded from image files as their diffuse channels, however (unlike in HW3) the chess pieces do not, as they use specialized materials. In particular, the chessboard shows very well that bilinear interpolation is working correctly for textures; it has notably lower resolution than the high-resolution render and clearly shows blurring between the spaces.

Looking at the lighting centered around the center of the board gives a clear indication that attenuation of the light sources is working as expected. The shadows also clearly encircle the center of the board, demonstrating clearly that their directionality is correct. The board itself also has a small reflective component to demonstrate that reflection on flat surfaces works correctly.

The white chess pieces have a diffuse component and a reflective component. By looking at the high- resolution render of this scene and zooming in on the pieces, the reflections can be easily seen and quite clearly demonstrate that barycentric interpolation of normals for triangles is working as intended.

The black chess pieces have a reflective component and a refractive component, along with a very small diffuse component to make them look more defined. They very clearly demonstrate the lensing effects of refraction through curved surfaces. They also demonstrate very nicely that the three different components (reflective, refractive, and diffuse) can be used together to create a very aesthetically pleasing effect. Note that their shadows are also much lighter than the white pieces, demonstrating partially opaque surfaces once again.

Take particular note of the chess pieces near the bottom of the image. The white pieces demonstrate multiple shadows due to the fact that this scene actually uses two different point light sources to get its lighting effect.

The complexity of this scene, with over 500,000 triangles in total, goes to demonstrate the sheer efficiency of the ray tracer in performing its renders. Even at very high resolutions, the rendering times for this scene are quite reasonable. As such, it acts as a good stress test of the ray tracer.

The 910-by-512 pixel image found in images/yours.ppm demonstrates this scene in very low quality. This render is simply meant to comply with the assignment directions and to include the output of the ray tracer as it was written to disk originally.

Two other renders can be found in images/8k-chessboard.png and images/chessboard-far.png. These images were converted from the PPM output of the ray tracer into PNG images for space reasons. The first demonstrates a very detailed render of the chessboard at a resolution of 8K (7680x4320) for the purposes of being able to zoom in very closely to see the details in the scene that may not be visible at lower resolutions; this render also uses 4x anti-aliasing (16 samples per pixel) to achieve maximum fidelity. For reference, this image took approximately 15 minutes to render.

The second of these images features a 1080p render of the chess scene from the perspective of the second camera defined in the scene to demonstrate that this functionality works correctly. It also shows much more clearly the shadows underneath the table and chairs and the large-scale attenuation of the light sources over long distances.

The scene itself can be found in the scenes/chessboard.scn file. The default camera was used for the first two renders and the far camera was used for the last.

Knight and Spheres

This scene is a very simple scene just meant to demonstrate more clearly that several features work as intended. Namely, the sphere in front of the camera demonstrates a particular type of distortion seen with refraction due to how the sphere acts like a lens. It also shows very clearly that ray depth is working correctly. By adjusting the depth and looking at the reflection seen through the refractive sphere, the effect of changing depth can be observed.

This scene also demonstrates that light colours can be changed and that multiple point light sources act as expected when their light mixes. The two light sources in this scene emit red and blue light, which can both be seen on the knight. Additionally, the reflective sphere is also partially diffuse and the specular highlights from both lights can be clearly seen in it.

A basic render of the scene in 1080p can be found in images/knight-and-spheres.png and the scene itself can be found in scenes/knight.scn.

Scene File Format

The scene file format used in HW3 has been slightly altered to allow the new features that are present in the ray tracer to be controlled. As before, each non-blank line specifies a command and all attributes for a command should be indented by the same amount and all colour values range from 0 to 1.

The following commands and attributes are available:

• The mdl <name> <obj file> command loads the given OBJ file into a model with the given name
• The mtl <name> command defines a new material with the given name.
  • The ambient <r> <g> <b> attribute defines the ambient reflectivity
  • The diffuse <r> <g> <b> attribute defines the diffuse reflectivity
  • The specular <r> <g> <b> attribute defines the specular reflectivity
  • The shininess <value> attribute defines the shininess exponent
  • The opacity <value> attribute controls the opacity of the object for lighting purposes
  • The diffuse_map <image> attribute defines the diffuse reflectivity map
  • The ao_map <image> attribute defines the ambient occlusion map
  • The reflectivity <value> attribute defines the reflectivity of the material
  • The transmittance <value> attribute defines the transmittance of the material
  • The refractive_index <value> attribute defines the index of refraction of the material
• The plight attribute defines a new point light (max 16 per scene)
  • The pos <x> <y> <z> attribute defines the position of the light within the scene
  • The ambient <r> <g> <b> attribute defines the ambient light intensity/colour
  • The diffuse <r> <g> <b> attribute defines the diffuse light intensity/colour
  • The specular <r> <g> <b> attribute defines the specular light intensity/colour
  • The atten <a0> <a1> <a2> attribute defines the attenuation coefficients
    • Light intensity is multiplied by 1 / (a0 + a1 * d + a2 * d * d) where d is the distance to the light
• The obj mesh command defines a new triangulated mesh object in the scene
  • The mdl <model name> attribute defines what model this object will use (required)
  • The mtl <material name> attribute defines what material this object will use
  • The pos <x> <y> <z> attribute defines the position of the object within the scene
  • The rot <yaw> <pitch> <roll> attribute defines the orientation of the object (rotations are performed about the object's origin and are specified in radians)
  • The scale <scale factor> attribute defines the object's scale factor (scaling is performed about the object's origin)
• The obj sphere command defines a new sphere object in the scene
  • The mtl <material name> attribute defines what material this object will use
  • The pos <x> <y> <z> attribute defines the center of the sphere
  • The scale <scale factor> attribute defines the radius of the sphere (the default sphere has a radius of 1)
• The cam <name> command defines a new camera (the camera with name default is used by default if no camera is specified on the command line)
  • The pos <x> <y> <z> attribute defines the location of the camera
  • The lookat <x> <y> <z> attribute defines the point to which the camera will look
  • The up <x> <y> <z> attribute defines the up vector of the camera
  • The hfov <value> attribute defines the horizontal FOV of the camera (in degrees)

Resources Used

• GLM Manual
• GLM API Reference
• Ray Tracing Refraction
• Möller-Trumbore Algorithm
• BVH Construction using Approximate Agglomerative Clustering
• Some code was loosely based on the skeleton code written by Karl Augsten
• Lots of code and scenes adapted from HW3