Description
You are tasked with building a ray tracer. Your ray tracer will output a single static PNG
image, based on an input ‘scene’ file and command line arguments. We have provided you
with a template C# implementation that you will need to complete. We are not using
the Unity engine in this project, however, you may find that some of the theory in
this assignment will be transferable to Unity development (particularly the maths). The
assignment is broken down into numerous stages and steps. Our expectations for each
stage and the respectively allocated marks are outlined in detail below. You should aim
to complete each step in sequence since this will make the process less overwhelming.
There are various approaches to modelling how light interacts with surfaces in a scene.
Almost always, the choice of approach comes down to a trade-off between computational
complexity and realism. A ray tracing based approach can produce very realistic images,
however this comes with a significant computational cost that generally makes it unsuit
able for real-time rendering. Even if there are no real-time rendering requirements, we
still have to approximate and optimise the ray tracing process, since simulating all rays
in a scene is computationally intractable.
Template code
You will be given a GitHub repository to work on your project that is already pre
initialised with the template code. This is a private repository, so you may commit/push
to it without worry of other students having access to your work. You are expected to
use GitHub from the start through to the end of the project, and should commit and
push frequently. We won’t accept submissions not hosted in your private repository.
ò
A link to accept the assignment and automatically create your template
repository is provided on the Canvas project page (where you found this
specification document). Note that you may submit the assignment as many
times as you wish – only the latest will be marked.
1Stage 1 – Basic ray tracer
You will first implement the basic functionality of a ray tracer. At its core, ray tracing
is an application of geometry and basic linear algebra (vector maths will become your
bread and butter!). For example, a ray of light can be modelled by two three-dimensional
vectors: a starting position and direction. Surfaces, light sources, and other entities
in the environment can also be defined using vectors. Using geometry, it is possible to
calculate how a ray reflects off a surface, or perhaps even refracts through it. Ultimately
we are interested in simulating rays of light propagating throughout the environment,
interacting with various surfaces, before finally reaching the viewer as pixels on their
screen. If we are clever in utilising ‘real-life’ physical models for these interactions, we
can generate incredibly realistic scenes.
In this first stage you will implement some basic vector functionality, and figure out
how to shoot a ray for each pixel in a rendered image. We won’t yet be worrying about
materials, lighting, shading, etc. Such fancy stuff will come later in the assignment.
Stage 1.1 – Familiarise yourself with the template
Before writing any code, try to understand how the template provided to you works. We
have already taken care of quite a few details for you, such as input and output handling.
A sample input scene is provided to you in a text file (tests/sample scene 1.txt), and
a parser for this file has been written so you can access objects and resources directly
within the Scene class (src/scene/Scene.cs). The core ray tracing logic (which you will
write) should be implemented inside the Render() method in Scene.cs. This method
takes an Image object for which you can set the individual colour of each pixel, as well
as derive properties such as its width and height. When the program is run, this image
will automatically be outputted as a PNG image file.
Try running the project so that you can see this in action. Open up the terminal in
Visual Studio Code (or your preferred environment), and run1:
dotnet run — -f tests/sample_scene_1.txt -o output.png
Although this looks like a bit of a mouthful at first, all it is doing is running the project
with two command line arguments: an input text file (-f) and an output image file
(-o). The input file will be read and parsed, and the output image written accordingly.
Open the generated output file, and you will notice the entire image is black, since no
ray tracing has been implemented yet. Before continuing, test your understanding by
modifying the project code to output the image entirely in white instead.
ò
Hint:
Try using some loops inside the
Render() method. The Image class
has
Width
and Height properties which should be handy for determining
the loop bounds. These properties are already determined by the command
line arguments -w and -h, if specified.
1Note that you may need to install the .NET SDK if you haven’t already, otherwise dotnet run won’t be
available. Make sure you install version 6.0. You can find it at https://aka.ms/dotnet-download.
2Now take a look at the main Program.cs file. In the OptionsConf class, you can see
all of the potential command line arguments and their default values (these are the values
used if that argument is not specified at runtime – e.g., not entered on the command
line). Don’t change these default values, instead, pass values using the appropriate
flags on the command line, if you want to change parameters. At this point it’s worth
stressing that you should not modify the Program.cs file at all. Doing so risks
our automated test suites breaking when running your project during marking (see the
‘Submission‘ section for details).
Stage 1.2 – Implement vector mathematics
We have provided you with a C# struct template for representing a three-dimensional
vector (src/math/Vector3.cs). Write code to complete the missing operations which
are currently empty methods. Note that for convenience we have overloaded operators2
such as +, *, /. This is a handy language feature that allows us to perform vector
arithmetic concisely:
Vector3 a = new Vector3(0, 1, 0);
Vector3 b = new Vector3(1, 1, 0);
Vector3 c = a + b; // We overloaded ‘+’ so c = (1, 2, 0)
As well as basic arithmetic, you will also need to complete functions to compute the
dot product and cross product (at least). The dot product will tell you how much two
vectors point in the same direction, and the cross product of two vectors will give you
the vector which is perpendicular to both of them (e.g., crossing the x- and y-axes would
give you the z-axis).
.
It is
strongly recommended that you test your implementations here thor
oughly. Vectors are utilised everywhere in this project, so a mistake here
can lead to a major headache down the line.
Stage 1.3 – Fire a ray for each pixel
We have already provided you with a ‘ray’ structure (src/math/Ray.cs). Notice that
it is simply a position (origin) and a direction, both represented as vectors. While it is
possible to trace rays forwards from light sources in the scene, it is far more efficient to
trace rays backwards from the camera. This is because most rays in the scene will never
be seen by the viewer, and computing these would be a waste of resources.
Inside the Render() method, write code to iterate through each pixel and construct a
corresponding ray that fires into the world. The biggest challenge here will be converting
from a two-dimensional pixel coordinate to a three-dimensional ray. You might find it
useful to consult external resources about the maths here.
2 https://docs.microsoft.com/en-us/dotnet/csharp/language-reference/operators/
operator-overloading
3In this project we want you to use a left-handed coordinate system. The camera should
be situated at the origin of the scene – (0, 0, 0) – looking forward along the positive
z-axis (‘into the screen’), with the positive x-axis pointing ‘right’, and the positive y-axis
pointing ‘up’. You should ensure that there is a horizontal field-of-view (FOV) of 60◦.
As a sanity check, a ray at the very center of the rendered image should point in the
direction (0, 0, 1). Rays at the corners of the image should have directions (±i, ±j, k),
where i = j if the image is square. You should ensure that your solution works when
different output image widths/heights are specified. For non-square images the vertical
FOV should vary to maintain the correct aspect ratio.
Stage 1.4 – Calculate ray–entity intersections
In this project a scene can contain three types of primitive entities – planes, trian
gles and spheres. If you haven’t already, open the template classes provided in the
src/primitives folder:
- Plane.cs – Represented by a point (center), and a vector representing the direc
tion it faces (normal – i.e., perpendicular to the actual surface of the plane). Note
this defines an ‘infinite’ plane.
- Triangle.cs – Represented by three points (v0, v1, v2). A clockwise winding
order defines the front face of the triangle.
- Sphere.cs – Represented by a point (center) and a radius.
All of these classes implement the interface SceneEntity. This means they all contain
a method called Intersect(), which takes a Ray as its input and returns a RayHit as
its output. The returned RayHit structure contains important information used during
ray tracing: the incident ray direction, the position of the hit, and the normal of the
surface at that position. It is also possible that there is no hit at all, in which case,
null should be returned instead of a RayHit instance. Your job is to implement the
Intersect() method for all three primitive entities. Again, you may find it helpful to
research common mathematical approaches to these problems if you are stuck.
Stage 1.5 – Output primitives as solid colours
You are finally ready to generate some graphical output! Earlier you computed a ray for
each pixel in the image. Extend this code to check for intersections with primitives in
the scene. You will need to make further additions to the Render() method. For each
ray, iterate through every entity in the scene and check whether there is an intersection
between the ray and the entity. If so, you should set the corresponding pixel colour to
the colour of the object. Ensure you correctly handle cases where there is more than
one entity that coincides with a ray.
In case your object-oriented programming is rusty, here is a template for looping
through all primitives/entities in the scene and checking if ray intersects with them:
4foreach (SceneEntity entity in this.entities)
{
RayHit hit = entity.Intersect(ray);
if (hit != null)
{
// We got a hit with this entity!
// The colour of the entity is entity.Material.Color
}
}
Note that entity is an interface, so we don’t know exactly which type of primitive it
is (plane, triangle or sphere), but that does not matter since we are only interested in
the intersection itself.
ò
We have provided you with sample outputs in the
images folder, so you have
an indication of how your output should look for stages 1 and 2 respectively.
Stage 2 – Lighting and materials
In this stage you will extend the ray tracer to handle lighting, and model different
types of materials. Some materials are more trivial to compute than others, and this
complexity ultimately boils down to how light rays interact with them.
Note that every entity is assigned a material. The material contains properties that
allow us to calculate how light interacts with the entity – for example, its colour, whether
it is opaque, reflective, transparent, etc. Open src/core/Material.cs to see our repre
sentation of a material in this project. Note that you already used the Color property
in the previous stage.
Stage 2.1 – Diffuse materials
We will first consider the case where a ray coincides with a diffuse surface which is
directly illuminated by a light source. When light hits an ‘ideal’ diffuse material, it scat
ters uniformly in all directions. This means it is viewer-independent, and the intensity
only varies depending on the angle of incidence between the light source and the surface.
Diffuse lighting is so trivial to compute that it is regularly used in real-time rendering
techniques (not just ray tracing).
In this stage you need to extend the ray tracer to handle materials with the Diffuse
type. Objects should be smoothly lit when this is implemented correctly. As a starting
point, take note of where you set the colour of a pixel currently. Instead of outputting
the material colour directly, you should compute it based on the following function:
C = (Nˆ ·
ˆ
L
)CmCl
…where Nˆ is the normal of the surface at the hit point,
ˆ
L
is the direction to the light
source from the hit point, Cm is the material colour, Cl is the light colour and C is the
5resultant output colour. Note that all light sources are available in the Scene class, so
you will likely have to iterate through these. You should sum the outputs of multiple
light sources into the final pixel colour (if there is more than one).
Stage 2.2 – Shadow rays
Consider the fact that light rays may be blocked by objects in the scene. This should
lead to visible shadows. Extend your implementation to check whether a hit point is in
fact in a shadow. You can do this by firing another ray towards the light source from
that point, and checking if it hits a (closer) surface along the way. If there is a hit, then
that light source should not contribute to illumination at that point.
.
Be careful when firing a ray away from the surface of an object. Numerical
error could lead to a ‘premature’ hit with that same object! One solution is
to offset the origin of the ray slightly away from the surface.
Stage 2.3 – Reflective materials
This is where ray tracing really starts to shine – no pun intended! Extend the ray tracer
to handle materials with the Reflective type. When a ray hits a reflective material,
another ray should be recursively traced to determine the colour at that point. To do
this you need to calculate a reflection vector as a function of the hit point’s surface
normal and the incident ray direction. This should be pure reflection – the colour of
the material plays no role in the calculations. Note that if there are a lot of reflective
surfaces in a scene, computational costs can blow out significantly, so you may wish to
place a hard limit on the depth of recursion (i.e., how many new ‘reflection’ rays you are
willing to ‘fire’).
Stage 2.4 – Refractive materials
Some materials are transparent, and allow light to transmit through them. Glass is an
example of such a material. Unfortunately, simulating this effect in a realistic manner
is not as simple as allowing an incident ray to pass directly through the object. Indeed,
you may have observed that light can ‘bend’ through transparent mediums (take a look
at any curved glass object). This phenomenon is known as refraction. Extend the ray
tracer once again to handle materials with the Refractive type. In a similar way to how
you handled reflection, upon a ray hitting a surface, you should recursively trace a ray
through the object according to physical laws of refraction. The colour of the material
should not play any role in the calculations at this stage. Note that materials have an
additional RefractiveIndex property, which will come in handy here.
Stage 2.5 – The Fresnel effect
In the real world, refraction does not really occur in total isolation from reflection. When
light hits a refractive surface, some proportion of it is reflected, while another proportion
6is refracted (these proportions sum to 1 since energy is conserved). This proportion is
not uniform for all rays which hit the surface. As a ray’s angle of incidence decreases,
there is greater reflection versus refraction. If you look at a sheet of glass from front on,
and you will see that most of the light is refracted (transmits through). However, if you
look at it almost side-on, it looks a lot more reflective!
This phenomenon is known as the Fresnel effect. Your next task is to improve re
fractive materials so that reflection is mixed into the corresponding lighting calculations
according to the Fresnel equations. Note that this means that two rays need to be traced
for every one ray that coincides with a refractive material. If this process repeats itself
multiple times, the computational burden increases exponentially, so keep this in mind
when coding your solution.
Stage 2.6 – Anti-aliasing
You may have noticed that the images being produced so far contain somewhat jagged
edges. This is because details in the scene can differ at the sub-pixel level when they
are projected onto the final image. This is a common problem in computer graphics
generally, and is called aliasing. Aliasing is usually quite visible where there are curved
edges, or edges that are not aligned horizontally or vertically with the screen (think
about why). We can use various techniques to mitigate this problem, and this process
is called anti-aliasing.
Modify your ray tracer to incorporate optional anti-aliasing during rendering. You
should do this by firing more rays per pixel and then averaging the outputs for the final
colour. There is another command-line argument accessible within your program which
specifies the anti-aliasing multiplier you should use when rendering the scene. Here is
an example:
dotnet run — -f tests/sample_scene.txt -o output.png -x 2
The argument -x specifies this multiplier, which in this example is 2. This means you
should fire twice as many rays both horizontally and vertically (4x rays per pixel). If the
multiplier is 3, then you should fire three times as many rays in both directions (9x rays
per pixel). And so on. Note that we have already parsed this command-line argument
for you! It is accessible within the Scene class as options.AAMultiplier, so you don’t
need to worry about how to read it into your program.
Stage 3 – Advanced add-ons
In this stage you are given the opportunity to implement some advanced add-on effects
of your choosing. Some are more trivial to implement than others, and the allocated
marks reflect their approximate difficulty and/or time commitment. In completing these
questions to a high standard, we expect that you research various approaches, and make
informed decisions to maximise the outcomes of the intended effects. You should write
some detailed comments in your README.md file which describe the approach you have
taken. It is not possible to receive more than the allocated marks for this section (9
7marks maximum), so if you complete more add-ons than required, clearly state which
ones you want to contribute to your mark!
N.B.: Regardless of whether you complete this stage or not, you are still encouraged
to show off your work by submitting a custom scene (see below).
Render quality
For a few of these add-ons, it may be useful to have a render quality parameter, which
allows the user to ‘trade’ between computation time and output image quality. We have
provided an additional command line argument -q that you are free to utilise for this.
For add-ons where it’s relevant, our test suite may increment the quality level and
re-render a test output (starting at the default quality level of 0). This will be repeated
until some reasonable time limit is reached, or the image does not change. The exact
details of this process will depend on available resources during marking, however all
submissions will be tested in the same environment, and with the same time limits, in
order to ensure fairness.
ò
You do not
have to utilise the -q parameter. It’s there purely so you have
the opportunity to show what your ray tracer is capable of without having
to be ultra-conservative with respect to time. It only applies to select stage
3 add-ons, and won’t be used elsewhere during testing. It is up to you to
determine if it is relevant to the problem you are solving.
Stage 3, Option A – Emissive materials (+6)
Up until this point we considered lights to be infinitely small points. This is an approx
imation, and not how lighting typically works in the real world. Consider a fluorescent
bulb which is a long cylindrical shape. It would be inappropriate to model this using
a singular point. Even a standard light globe, which comes close to being modelled by
a ‘point’, is still a physical object with a surface area that emits light. We call such
surfaces emissive.
For this add-on you need to extend the ray tracer to handle emissive materials. As
a consequence there should also be ‘soft shadows’ present in the scene (have a think
about why). Note that the MaterialType enum has an Emissive type, which will be
used to specify whether the material should be emissive. The colour of the material can
be interpreted to be the colour of the emitted rays. You may assume only spheres and
triangles will be given emissive materials (not infinite planes).
Stage 3, Option B – Ambient lighting/occlusion (+6)
The current ray tracer only simulates a very small subset of rays in a scene. Consider
how the illumination of a point on a diffuse object is currently calculated. We test
if there is a direct ray originating from a light source. If there isn’t, or there is an
object in-between, then the point isn’t illuminated at all. However, in reality, there
may still be some illumination that comes from indirect rays of light (e.g. bouncing off
8other surfaces in the scene). This is called ambient lighting. When ambient lighting is
computed, we consequently compute ambient occlusion in the scene, since indirect light
will not illuminate all surfaces equally. For example, surfaces inside crevices and cracks
tend to be a lot darker since they are less ‘exposed’ to the other surfaces in the scene.
Extend the ray tracer so that it is possible to optionally compute ambient lighting
in a rendered image. You should utilise the -l command line flag which is available as
options.AmbientLightingEnabled in the Scene class. In other words, if and only if
-l is passed to your program, ambient lighting should be enabled.
ò
It is infeasible to compute
every ray in a scene, so you may end up using
some sort of randomised or ‘Monte Carlo’ method to assist with this.
Stage 3, Option C – OBJ models
Extend the ray tracer to read simple 3D models in the form of .obj files. Research how the
.obj file format is structured. We only expect that you handle files with vertex, normal
and face definitions (v, vn, f), and faces may be assumed to be triangles (exactly three
vertices). Be sure to consider how more complex models could impact the performance of
your ray tracer, and optimise relevant data structures accordingly. If you do implement
optimisations, ensure these are discussed in your README.md.
We have provided you with a template class in the src/extensions folder called
ObjModel.cs. In the same manner as the sphere, plane and triangle primitives defined
earlier, ObjModel has an Intersect() method that needs to be implemented. You
should read/parse the .obj file in the constructor, and use this data when computing
intersections.
We have provided you with a sample scene (tests/sample scene obj.txt) that con
tains an OBJ, so you should examine this file first to see exactly how OBJ models are
defined. In particular, observe there is a string that is the path to the .obj file itself.
Also notice that we specify an offset vector and scale as parameters when loading a
model. These parameters allow us to adjust the size and position of the model to fit the
scene appropriately. The scale should be applied first, then the offset, as these are order
dependent operations.
Stage 3, Option D – Glossy materials
At the moment we can model a few common materials, but there are still many ma
terial types which cannot be adequately simulated. Add support for ‘glossy’ materials
to your ray tracer. The MaterialType enum has a Glossy type, which you should add
functionality for. Unlike refractive materials, no light should transmit through the ob
ject. However, the object should not be completely reflective either. Instead, the object
should be coloured based on the material’s Color property, but also appear to reflect
some light. Exactly how you achieve this effect is up to you, but you should aim to make
it convincing. If there are parameters that need to be tuned to determine the amount
of ‘glossiness’, choose suitable defaults that demonstrate the effect.
9Stage 3, Option E – Custom camera orientation
Currently the camera is assumed to be at the origin (0, 0, 0) looking along the z-axis
(with the y-axis oriented ‘up’). Allow the user to apply a custom camera position and
orientation to the camera based on the following parameters: a position vector (x, y, z),
an axis of rotation vector (x, y, z) and an angle (in degrees). Note that this is an axis
of rotation, not the ‘direction’ or local z-axis of the camera. The axis and angle are
specified via the following command-line parameters where # denotes a floating point
number:
- –cam-pos #,#,# – Available as options.CameraPosition
- –cam-axis #,#,# – Available as options.CameraAxis
- –cam-angle # – Available as options.CameraAngle (in degrees)
These parameters are automatically parsed for you by the template code, and so
all you need to do is utilise the respective variables within the Scene class (e.g., as
options.CameraPosition).
Stage 3, Option F – Beer’s law
We currently consider transmission of light through a medium to be unaffected by that
medium, other than its refractive index. Beer’s law states that light intensity should
drop as light passes through a medium, and that drop is proportional to the distance
traveled. Additionally, some wavelengths of light may be absorbed faster than others,
giving rise to ‘coloured’ semi-transparent objects.
Modify the ray tracer to obey Beer’s law for Refractive materials. Currently the
colour of refractive materials has no effect in lighting calculations, but now it should be
utilised. You should interpret this field to contain the absorbance of each colour channel.
For example the RGB value of (4, 1, 1) means that red light will be absorbed at a faster
rate than blue and green light.
Stage 3, Option G – Depth of field
The camera is currently modelled assuming an infinitely small ‘pinhole’ aperture. If
you think about it, a physical camera cannot function like this, since no light can pass
through an infinitely small hole! We may introduce additional parameters to our ray
tracer that allow us to more accurately model how a real camera works. This allows us
to generate a ‘depth of field’ effect in our rendered images, which means some parts of
the scene are in focus, and other parts are blurry depending on a given ‘focal length’.
Extend your ray tracer to account for two additional parameters: a focal length, and
an aperture radius. These will be passed via the following command-line parameters
where # denotes a floating point number:
- –aperture-radius # – Available as options.ApertureRadius
10• –focal-length # – Available as options.FocalLength
We have already parsed these parameters for you and you can access them in the
Scene class (e.g. as options.FocalLength). The aperture radius parameter has a
default value of 0, which reflects the idealised pinhole model we have used up until this
point. However, if this value is larger than 0, then a circular pinhole should be simulated
with the corresponding radius. Surfaces at a distance equal to the focal length (away
from the camera) should be in focus.
Finally – Show off your work!
y tracer should be able to generate some pretty impres
sive images. Now is your chance to show off what it can do. The tests/final scene.txt
file is currently empty. You should populate this scene in order to demonstrate all the
core features of the ray tracer, as well as your chosen add-ons. Choose suitable settings
to render your final image, and replace the images/final scene.png placeholder image
with it. Notice that this image is already embedded in the README.md.
Also record how long it takes to render the image on your PC and include that in
the README.md (see the template provided). In case we need to verify your image, you
must also include the exact command line used to generate it. We might use lower
resolution/anti-aliasing/quality settings to speed up the verification process.
.
Verify that your image is visible on GitHub (instead of the placeholder
image) to ensure that you receive credit for it.
