Feng Liu, PhD
NVIDIA has awarded Feng Liu, PhD, assistant professor of computer science at Drexel University’s College of Computing & Informatics, an NVIDIA Academic Grant Program award to support his research in autonomous driving AI.
The grant, in the form of computing resources of 32,000 A100 GPU-hours via NVIDIA’s cloud platform acquisition Brev, provides Feng and his research team access to a high-performance server equipped with eight NVIDIA A100 80GB GPUs.
“Think of it as a ‘high-end AI training machine’ that lets us run large-scale experiments,” said Liu.
Liu’s team will focus on developing advanced image-generation models that produce photorealistic driving scenes while preserving geometric consistency, including accurate camera viewpoints, vehicle placement and depth/3D structure.
“Geometric accuracy matters in autonomous driving,” said Liu. “Realism alone is insufficient; models must also be spatially correct.”
The project aims to build a pipeline that transforms scenario descriptions into structured scene plans, derives geometry-aligned ground-truth signals from 3D environments and trains diffusion models to generate realistic images that adhere to those planned geometric constraints.
“The resulting capability is intended to support the creation of synthetic training data for rare but safety-critical conditions (fog, night, heavy rain, unusual traffic configurations) that are difficult and costly to capture at scale in real-world data collection, helping improve robustness of perception systems for autonomous driving,” said Liu.