2019/10/02 by Alexander Wong, Mahmoud Famuori, Wong, Alexander +9 · 1 citation
Computer Science · Engineering · Medicine · #Advanced Neural Network Applications #COVID-19 diagnosis using AI #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Industrial Vision Systems and Defect Detection #Neural and Evolutionary Computing (cs.NE)
paper · pdf · doi:10.48550/arxiv.1910.01271
openalex publication_date 2019/10/02 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Object detection remains an active area of research in the field of computer\nvision, and considerable advances and successes has been achieved in this area\nthrough the design of deep convolutional neural networks for tackling object\ndetection. Despite these successes, one of the biggest challenges to widespread\ndeployment of such object detection networks on edge and mobile scenarios is\nthe high computational and memory requirements. As such, there has been growing\nresearch interest in the design of efficient deep neural network architectures\ncatered for edge and mobile usage. In this study, we introduce YOLO Nano, a\nhighly compact deep convolutional neural network for the task of object\ndetection. A human-machine collaborative design strategy is leveraged to create\nYOLO Nano, where principled network design prototyping, based on design\nprinciples from the YOLO family of single-shot object detection network\narchitectures, is coupled with machine-driven design exploration to create a\ncompact network with highly customized module-level macroarchitecture and\nmicroarchitecture designs tailored for the task of embedded object detection.\nThe proposed YOLO Nano possesses a model size of ~4.0MB (>15.1x and >8.3x\nsmaller than Tiny YOLOv2 and Tiny YOLOv3, respectively) and requires 4.57B\noperations for inference (>34% and ~17% lower than Tiny YOLOv2 and Tiny YOLOv3,\nrespectively) while still achieving an mAP of ~69.1% on the VOC 2007 dataset\n(~12% and ~10.7% higher than Tiny YOLOv2 and Tiny YOLOv3, respectively).\nExperiments on inference speed and power efficiency on a Jetson AGX Xavier\nembedded module at different power budgets further demonstrate the efficacy of\nYOLO Nano for embedded scenarios.\n