Benchmarking embedded deep learning hardware for computer vision

• Main Author: Ching, Amos Li En
• Other Authors: Mohamed M. Sabry Aly
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision; Engineering::Electrical and electronic engineering::Computer hardware, software and systems
• Online Access: https://hdl.handle.net/10356/138345
• Institution: Nanyang Technological University

The rapid rise of artificial-intelligence (AI) applications on big data such as image collection, has triggered a growing interest for companies to design ASICs (application-specific integrated circuit) for various AI applications. ASICs designed for AI application are playing an important role due to the huge advantages over general purpose processors such as CPUs. As demand of AI on edge computing grows, more embedded products are being designed with power efficient in mind. However, due to the different technologies used by companies, technical specifications cannot be easily compared. Thus, studies are needed to determine the power efficiency and inference performance. 3 edge platforms will be tested: NVIDIA Jetson, Intel Neural Compute Stick 2 and Qualcomm Snapdragon 835. Results will be compared with desktop-grade GPU power efficiency to determine the advantages of embedded devices for edge applications. Pre-trained Objection Detection models such as Faster R-CNN and SSD will be used for the benchmarks.