“To Infinity and Beyond: a GPU-Driven Memory Sharing Pipeline to Generate and Process Infinite Synthetic Data” by Pietra, Hahn and Garau
Conference:
Type(s):
Title:
- To Infinity and Beyond: a GPU-Driven Memory Sharing Pipeline to Generate and Process Infinite Synthetic Data
Session/Category Title:
- Rendering
Presenter(s)/Author(s):
Abstract:
Our GPU-resident pipeline based on Unreal Engine 5 unifies scene generation, rendering, and processing entirely on the GPU, eliminating CPU–GPU transfers and disk I/O to achieve near-constant per-sample latency, up to 12× speedups, and sustained high-throughput training with effectively infinite synthetic data.
References:
[1] Thomas Pollok, Lorenz Junglas, Boitumelo Ruf, and Arne Schumann. 2019. UnrealGT: Using Unreal Engine to Generate Ground Truth Datasets. In Advances in Visual Computing : 14th International Symposium on Visual Computing, ISVC 2019, Lake Tahoe, NV, USA, October 7–9, 2019, Proceedings, Part I. Ed.: George Bebis. Springer, 670–682.
[2] Shital Shah, Debadeepta Dey, Chris Lovett, and Ashish Kapoor. 2017. AirSim: High-Fidelity Visual and Physical Simulation for Autonomous Vehicles. In Field and Service Robotics. arXiv:arXiv:1705.05065https://arxiv.org/abs/1705.05065
