Technical Name Lightweight AI inference engine for smart LiDAR
Project Operator National Yang Ming Chiao Tung University
Project Host 郭峻因
Summary
This technology specializes in low-latency, low-power smart LiDAR and edge AI systems. By employing advanced lightweight model training and model caching techniques, it effectively reduces bandwidth requirements during AI inference. Notably, when processing the lightweight YOLOv7-Tiny model, it achieves an impressive 4500 FPS while consuming only 38mW of power.
Scientific Breakthrough
By leveraging advanced lightweight model training and model caching technologies, this technology significantly reduces bandwidth requirements for AI inference. Furthermore, with support for multi-scale dynamic fixed-point computation, it ensures high accuracy even when utilizing extremely lightweight models, making it an ideal solution for edge AI applications.
Industrial Applicability
This technology delivers key advantages, including low latency and low power consumption, making it an optimal choice for integration into edge AI systems. It is particularly well-suited for applications demanding rapid response, such as high-speed aerial vehicles and automobiles, as well as for systems requiring prolonged operation, such as surveillance and monitoring solutions.
  • Contact
  • Peggie Chen