Database Portfolio
Drawing Project
Floating Island
Jeremy The Rat
MaebeeTV Discord Bot
Maya Forest Generator
Super-pixel Importance Sampling
Sweet N Soft tech
The Forest
Vernos Langathripoodinary IIISuper-pixel Importance Sampling
Created on: 13/05/2024
Worked on by: Thomas Lower
This project is based on mathematics from my Advanced Maths unit from Bournemouth University.
The end goal of the project was to produce a realistic renderer capabale of rendering mathematically accurate spheres.
Sampling Types
The program currently supports 3 main types of sampling:
Linear Sampling
The Linear Sampling algorithm simply works by sampling each pixel a set number of times, and repeating this for every pixel in the window. This provides the most accurate results at the cost of a lack of time efficiency as pixels are not judged on whether they are needed to be rendered.
A render from the raytracer using linear sampling with 1 sample per pixel, 1 reflection bounce and 1 real pixel per data pixel
A render from the raytracer using linear sampling with 3 samples per pixel, 1 reflection bounce and 1 real pixel per data pixel
A render from the raytracer using linear sampling with 1 sample per pixel, 3 reflection bounces and 1 real pixel per data pixel
A render from the raytracer using linear sampling with 1 sample per pixel, 1 reflection bounce and 5 real pixels per data pixel
This system also works the least well with super-pixels (virtual pixels that take up the space of multiple real pixels). This is due to the fact that pixels are generated every 1 * pixelSize pixels meaning that superpixels cannot overlap.
Randomised Sampling
The Randomised Sampling algorithm works by randomly generating a pixel, this is much faster at the cost of less clear images
A render from the raytracer using randomised sampling with 1 sample per pixel, 1 reflection bounce, 1 real pixel per data pixel and 500 random passes
Importance Sampling
The Importance Sampling algorithm is based on the randomised sampling algorithm, but performs multiple passes at different pixel sizes, avoid pixels that are found to be empty in previous passes.
A render from the raytracer using importance sampling with 1 sample per pixel, 500 passes and 1 real pixel per data pixel
A render from the raytracer using importance sampling with 3 sample per pixel, 500 passes and 4 real pixel per data pixel
Interactive rendering
The following videos demonstrate the renderer performing importance sampling over time.
A video of the raytracer rendering with importance sampling
A video of the raytracer rendering with importance sampling with a minor fix to improve the quality of renders
Graph of Image Readability over time
This graph demonstrates how the comprehendability of the image increases over time.
A graph of image readability over time
The Graph demonstrates that a linear sampler is only useful once completed as before this, it does not contain data from the entire image, just a fraction. The randomised sampler provides much better data part of the way through, however, can still be unclear towards the beginning. My importance sampling algorithm provides the best data early on however, this eventually plateus so that the image is never truly complete.
See for yourself
The code for this engine can be found on Github. To run this you will need SDL2 and SDL2-image installed as well as g++. It can be built and ran by running the Makefile.