View, browse and sort the ever-growing list of sessions by pass type, topic, format and viewing experience. Use the experience filter to view sessions and roundtables that will be available on the virtual GDC event platform. Virtual sessions will be available to watch on-demand through April 1st, 2022 at 5pm PT. All days/times are displayed in Pacific Time.
Use the Viewing Experience filter to see which sessions will be available on-site, on the virtual platform, or both:
All registered attendees will be able to build their personal schedule in the virtual GDC event platform and mobile app. Download the Informa Tech Events app to gain access to the Game Developers Conference 2022 virtually. Login with the email address you used to register.
Thomas Müller (NVIDIA)
Location: Virtual GDC Platform
Date: Thursday, March 24
Time: 12:45 pm - 1:15 pm
Pass Type: All Access Pass, Core Pass, Summits Pass, Expo Plus Pass, Audio Pass, Independent Games Summit Pass, All Access Online Only Pass - Get your pass now!
Format: Sponsored Session
Viewing Experience: Virtual
Vault Recording: Video
Audience Level: Yes
Neural representations of computer graphics primitives, such as light fields, volumetrics (NeRF), signed distance functions (SDFs), and the like, have dramatically risen in popularity in recent years. However, despite advances in fast rendering of such primitives, their efficient training remains an elusive goal, taking typically on the order of hours. We'll show how such neural graphics primitives can be trained in seconds rather than hours, as well as rendered at faster-than-real-time rates (>60fps >1080p) without any post-processing of the trained model. We'll focus on three key ingredients, which combined can deliver a more than 1,000x speedup (in some cases) compared with prior work.
The session teaches two ingredients: (i) a multiresolution hash encoding and (ii) fully fused neural networks, which, when combined, allow training neural graphics primitives such as radiance caches, NeRFs, and SDFs within seconds and rendering them faster than real-time rates (milliseconds).
Programmers interested in applying neural networks to real-time graphics and aren't afraid of low-level implementation details such as those of training and running neural networks.