Dongxu Lyu
Papers
2
Total Citations
12
H-Index
2
About
Dongxu Lyu is a researcher at the forefront of efficient 3D perception hardware, specializing in computer architecture and accelerator design for point-cloud processing. His primary research focuses on developing specialized hardware accelerators for 3D sparse convolution (SpConv) networks, which are critical backbones for applications in robotics, autonomous driving, and augmented/virtual reality. Lyu’s major contribution is the design of SpOctA, a novel 3D sparse convolution accelerator that introduces octree-encoding-based map search and inherent sparsity-aware processing. This innovation directly addresses the severe computational and memory bottlenecks posed by irregular, sparse point-cloud data, enabling faster and more energy-efficient 3D perception. While his most-cited work has garnered over 10 citations, reflecting its timely relevance, Lyu’s impact lies in pushing the boundaries of efficient on-device AI for real-world 3D vision tasks. His work is particularly notable for bridging algorithmic sparsity with hardware efficiency, a crucial step toward deploying advanced perception systems in resource-constrained edge devices. As a rising voice in the field, Lyu’s research promises to accelerate the adoption of 3D deep learning in safety-critical and interactive applications.
Research Focus
Key Achievements
Top Papers
- 1
- 2