YOLO-LITE: A Real-Time Object Detection Algorithm Optimized for Non-GPU Computers

TLDR

YOLO‑LITE builds on YOLOv2 to provide a smaller, faster, and more efficient real‑time object detection model, expanding accessibility to devices without GPUs. The study aims to develop YOLO‑LITE so that real‑time object detection can run on portable devices such as laptops and cellphones lacking GPUs. YOLO‑LITE achieves this with a lightweight 7‑layer architecture requiring only 482 M FLOPS, enabling inference at 21 FPS on a non‑GPU computer and 10 FPS when deployed on a website. The model attains 33.81 % mAP on PASCAL VOC and 12.26 % on COCO, runs at 21 FPS on non‑GPU hardware and 10 FPS on a web deployment, and is 3.8× faster than the fastest state‑of‑the‑art SSD‑MobilenetV1.

Abstract

This paper focuses on YOLO-LITE, a real-time object detection model developed to run on portable devices such as a laptop or cellphone lacking a Graphics Processing Unit (GPU). The model was first trained on the PASCAL VOC dataset then on the COCO dataset, achieving a mAP of 33.81% and 12.26% respectively. YOLO-LITE runs at about 21 FPS on a non-GPU computer and 10 FPS after implemented onto a website with only 7 layers and 482 million FLOPS. This speed is 3.8 × faster than the fastest state of art model, SSD MobilenetvI. Based on the original object detection algorithm YOLOV2, YOLO-LITE was designed to create a smaller, faster, and more efficient model increasing the accessibility of real-time object detection to a variety of devices.