Rise Again

Implementation

---

Triangle Ring

This is by far the most prominent light source in the render. The triangle ring is an additional solid I implemented on top of the regular triangle. It uses barycentric coordinates to compute the distance between the hit point and the outer edge of the triangle. If that distance is greater than a given thickness parameter, the intersection is ignored.
The solid also implements the area and sample functions needed to use it as an area light. I used two of these triangle rings with opposite orientations stacked on top of each other to allow the ring to cast light both upwards and downwards. To prevent intersections with the wrong solid, the triangle ring is completely invisible from one side.
By far the biggest challenge with the triangle ring was getting the part of the statue, which is on the same plane as the triangle to converge reasonably well. Because of the steep angle to the area light, most light samples from this "belt region" are either dark or occluded. Only very few samples actually contribute to the final color, which leads to extremely slow conversion. To fix this, the surface normal returned by the sample function is slightly randomly perturbed, which effectively blurs away the problematic region. To get rid of any remaining artifacts, a simple star shaped median filter is applied to the image, once the render is complete.

Code path: rt/solids/triring.cpp, main/a_compo.cpp @ LN 198

God Rays

The renderer implements a simplified version of volumetric rendering to give more focus the spot light in the center of the scene. This was done by implementing a new "Holy Recursive Ray Tracing Integrator" which extends an integrator from a pervious assignment by an additional volumetric pass.
This pass works by intersecting a ray with the scene, then walking along the ray and checking for visibility with the spotlight in each step. The samples checked along the ray are at equal distance to each other, but offset by a random amount each time. Averaging over these samples gives us an approximation of how much light is being scattered towards the camera.
In order to make the spotlight a little bit more natural, I implemented a new "Area Spotlight", which randomizes it's position within a rectangular region. This helps to feather out the outer edge of the cone, in which the spotlight functions.

Code path: rt/integrators/holyrecraytrace.cpp, rt/lights/areaspotlight.cpp

Tesselation

To make use of the "SmoothTriangle" class I implemented for a previous assignment, I added a "tessel" function, which takes the triangles created by the "loadOBJ" function as input and generates smooth triangles that represent the same geometry.
The function first iterates over the list of triangles and collects all vertices in the geometry as well as the normals of the triangles, which are connected to those vertices. In a second pass, the flat triangles are swapped out for smooth triangles, if the angle between the vertex normals is small enough. If two of the vertex normals point too far away from each other, interpolating between those normals may result in a zero vector. In this case, the triangle is not replaced.

Code path: main/a_compo.cpp @ LN 243

Bloom Pass

The bloom effect is a tool to make the highlights of the scene stand out even more. Something like this happens with real cameras when parts of an image are overexposed and the brightness bleeds onto the surrounding pixels. This effect is applied after the ray tracer finished rendering the image in it's entirety and before color correction is applied. In software this effect can be implemented efficiently using an iterative algorithm:

1. Make a copy of the image produced by the ray tracer
2. Lower the blacks to ensure only the highlights of the image contribute to the bloom effect
3. Downscale the copied image by a factor of 2
4. Blur the copied image
5. Add the blurred image back to the original image
6. Go back to step 3 and repeat a couple of times

Since the image we need to blur in each iteration quarters in size every time, this algorithm is extremely fast, making it suitable even for real-time applications.

Code path: main/a_compo.cpp @ LN 57

Post Processing

The post processing pass is the last alteration of the image before it is written into the final png file. Its main propose is color correction, but some color grading is also applied. This pass consists of many small operations which are applied to each pixel in the image. I'm going to summarize these steps briefly here:

• First the reinhardt operator is used to map the value of each color from [0, ∞) down to [0, 1)
• Then brightness and saturation are raised back up a little bit
• A smoothstep curve is used to add more contrast to the image
• Now gamma correction is applied
• A bit of film grain is added to reduce color banding

Code path: main/a_compo.cpp @ LN 145