Deploying machine learning models to edge devices has many real-world applications, especially for the scenarios that demand low latency, low power, or data privacy. However, it requires substantial research and engineering efforts due to the limited computational resources and memory of edge devices. In this demo, we present BED, an object detection system for edge devices practiced on the MAX78000 DNN accelerator. BED integrates on-device DNN inference with a camera and a screen for image acquisition and output exhibition, respectively. Experiment results indicate BED can provide accurate detection with an only 300KB tiny DNN model.
翻译:向边缘装置部署机器学习模型有许多实际应用,特别是在需要低延迟、低功率或数据隐私的假设情景方面。然而,由于计算资源和边缘装置的内存有限,这需要大量的研究和工程努力。在此演示中,我们介绍BED,这是在MAX78000 DNN加速器上练习的边缘装置的物体探测系统。BED将DNN推论与摄像头和图像获取和输出展览的屏幕结合起来。实验结果显示BED只能提供300KB小DNN模型的精确探测。