Authors:
Abstract: The Deep Learning inference applications are moving to the edge side, as edge-side AI platforms are cheap and energy-efficient. Different edge AI processors are diversified, since these processors are designed with different approaches. However, it is hard for customers to select an edge AI processor without an overall evaluation of these processors. We propose a three-dimensional characterization (i.e., accuracy, latency, and energy efficiency) approach on three different kinds of edge AI processors (i.e., Edge TPU, NVIDIA Xavier, and NovuTensor). We deploy YOLOv2 and Tiny-YOLO, which are two YOLO-based object detection systems, on these edge AI platforms with Microsoft COCO dataset. I will present our work starting from the problem statement. And then I'll introduce our experiments setup and hardware configuration. Lastly, I'll conclude our experimental results and current work status, as well as the future work.
Best Poster Finalist (BP): no
Poster: PDF
Poster summary: PDF
Back to Poster Archive Listing