In the last decade, machine learning has become increasingly prevalent due to leaps in technology such as the invention of GPUs that are acceleration enabled for Neural Networks. These advancements have opened new doors for real-time applications to benefit from machine learning. However, Neural Networks and other machine learning techniques are not designed to run at the speeds required for fast, realistic rendering in high-performance environments. There is research being done to adapt machine learning approaches for a more complex replication of a real-world scene, such as denoising images and upsampling them to create crisper visuals.
This thesis looks at two distinct methods which seek to improve the realistic renderings of a scene with the aid of machine learning. Paper I explains the use of a neural network application which can compress light fields onto spheres in order to make them look more realistically illuminated in real-time environments. Paper II on the other hand, illustrates the use of a filtering method powered by a small convolutional neural network capable of denoising hair in real-time with stochastic transparency.
The person mentioned in this article is likely the author of the thesis. The author has looked at the use of machine learning to improve the realistic renderings of a scene. The author has also examined the use of a neural network application to compress light fields onto spheres to create more realistic illumination in real-time environments. Furthermore, the author has explored a filtering method powered by a small convolutional neural network capable of denoising hair in real-time with stochastic transparency.
The company mentioned in this article is likely the company that provided the machine learning resources used by the author in the thesis. The company has provided the author with the powerful applications necessary to explore the full potential of machine learning. With their resources, the author was able to explore new uses of machine learning such as the creation of a neural network application and a filtering system to power new, realistic renderings of a scene. The company has allowed for the author to take the necessary steps to bridge the gap between machine learning and real-time applications.
