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

**Authors:** Jonathan Pedoeem, Rachel Huang  
**arXiv:** [1811.05588](https://arxiv.org/abs/1811.05588)  
**Date:** 2018-11-14  
**Summary page:** https://www.summarizepaper.com/en/paper/1811.05588v1/

## Key points

- YOLO-LITE is a real-time object detection algorithm optimized for non-GPU computers
- It is designed to run on portable devices such as laptops or cellphones without a GPU
- YOLO-LITE achieved mean Average Precision (mAP) scores of 33.81% and 12.26% on the PASCAL VOC and COCO datasets respectively
- It runs at approximately 21 frames per second (FPS) on a non-GPU computer and 10 FPS on a website
- YOLO-LITE is 3.8 times faster than the fastest state-of-the-art model, SSD MobilenetvI
- The authors provide detailed information about the architecture and implementation of YOLO-LITE, including its seven layer design and computational efficiency
- The paper includes additional context about the affiliations of Rachel Huang, Jonathan Pedoeem, and Cuixian Chen
- YOLO-LITE increases accessibility to real-time object detection across various devices due to its smaller size and higher speed compared to existing models

## Summary

This paper presents YOLO-LITE, a real-time object detection algorithm optimized for non-GPU computers. The model is designed to run on portable devices such as laptops or cellphones that lack a Graphics Processing Unit (GPU). YOLO-LITE was trained on the PASCAL VOC and COCO datasets, achieving mean Average Precision (mAP) scores of 33.81% and 12.26% respectively. It runs at approximately 21 frames per second (FPS) on a non-GPU computer and 10 FPS when implemented on a website with only 7 layers and 482 million Floating Point Operations Per Second (FLOPS). This speed is 3.8 times faster than the fastest state-of-the-art model, SSD MobilenetvI. Based on the original YOLOV2 object detection algorithm, YOLO-LITE aims to create a smaller, faster, and more efficient model to increase the accessibility of real-time object detection across various devices. The authors provide detailed information about the architecture and implementation of YOLO-LITE including its seven layer design and computational efficiency. Furthermore, they include additional context from Rachel Huang's affiliation with the School of Electrical and Computer Engineering at Georgia Institute of Technology in Atlanta, United States; Jonathan Pedoeem's affiliation with Electrical Engineering at The Cooper Union in New York; and Cuixian Chen's affiliation with Mathematics and Statistics at UNC Wilmington in North Carolina. Overall, this paper contributes to the field of computer vision by introducing YOLO-LITE as an optimized solution for real-time object detection on non GPU computers which increases accessibility across various devices due to its smaller size and higher speed compared to existing models.