Practical Guide to Implementing Deep Neural Network Inferencing at the Edge


Date:  Thursday, September 17, 2020

Start Time: 10:00 am

End Time: 10:30 am

In this presentation, we explore practical aspects of implementing a pre-trained deep neural network (DNN) inference on typical edge processors. First, we briefly touch on how we evaluate the accuracy of DNNs for use in real-world applications. Next, we explain the process for converting a trained model in TensorFlow into formats suitable for deployment at the edge and examine a simple, generic C++ real-time inference application that can be deployed on a variety of hardware platforms. We then outline a method for evaluating the performance of edge DNN implementations and show the results of utilizing this method to benchmark the performance of three popular edge computing platforms: The Google Coral (based on the Edge TPU), NVIDIA Jetson Nano and Raspberry Pi 3.

Track

Session Speakers

  • Toly Kotlarsky is a pioneer of smart cameras and imaging barcode scanning technologies. He has authored real-time operating systems and kernels; designed system architectures and led the development of highly successful commercial products. Since the mid-1990s, the focus of his career has been in the development of 2D image processing systems. At the startup company Auto Image Id, Inc., where he was Director of Software Engineering, he and his team developed the world's first industrial smart camera. In 2003, Kotlarsky joined Metrologic Instruments, Inc., later acquired by Honeywell, to lead their Advanced Development R&D group and run the program of development of imaging technologies. The company's first imaging barcode scanner, called Focus, was developed by his team and released to production in May of 2005. It was the world's first imaging barcode scanner adopted by a top retail chain, which had transformed the entire retail industry. Currently, he is proudly working for Zebra Technologies, a great technology company.


Share

Practical Guide to Implementing Deep Neural Network Inferencing at the Edge