Stay Informed:
Baskin Engineering COVID-19 Information and Resources
Campus Roadmap to Recovery
Zoom Links: Zoom Help | Teaching with Zoom | Zoom Quick Guide

Advancement: Enabling Next-Gen Real-time 3D Graphics with Machine Learning

Speaker Name: 
Manu Mathew Thomas
Speaker Title: 
PhD Student
Speaker Organization: 
Computational Media PhD
Start Time: 
Thursday, May 20, 2021 - 11:00am
End Time: 
Thursday, May 20, 2021 - 12:00pm
Zoom - - Passcode: 951704

Abstract: Video game graphics are seeing a shift towards more realistic effects with the recent advancement of ray tracing hardware. Due to the real-time constraints of a game, we can only perform these effects in a limited capacity causing visual artifacts such as aliasing and noise. Machine learning-based approaches have shown promise in offline rendering but are expensive for games and other interactive media. One solution is to quantize the neural networks to use reduced-precision arithmetic, as it greatly improves their storage and computational costs. However, using a quantized network for the reconstruction of HDR images can lead to a significant loss in image quality. In our previous work, we introduced QW-Net, a neural network for image reconstruction, in which close to 95\% of the computations can be implemented with 4-bit integers. We then applied this network to solve the aliasing problem in video games, and our network produces significantly better quality than TAA, a widely used technique in current games. In this advancement talk, we propose 1) an extension to QW-Net for denoising and 2) A unified network framework for combining antialiasing, denoising, and super-resolution into a single network pass.

Event Type: 
Angus Forbes
Graduate Program: 
Computational Media PhD