Performance Analysis of the Pretrained EfficientDet for Real-time Object Detection on Raspberry Pi

Vidya Kamath, A. Renuka

Research output: Chapter in Book/Report/Conference proceedingConference contribution

4 Citations (Scopus)

Abstract

Recently there has been a lot of demand for deep learning models that can operate on a constrained device. When it comes to the task of object detection, EfficientDet is a well-known model. In this study, we use the integer quantization technique to perform real-time object detection on a Raspberry Pi using the popular EfficientDet family. We use the pretrained models from the TensorFlow to perform object detection for a specific task and evaluate their on-device performance. We examined the models' performance in terms of average precision and recall, IOU, speed, and model size. When working on a Raspberry Pi, we discovered that EfficientDet1 after quantization, with a moving average decay of 0.95 and a Stochastic Gradient Descent optimizer is a good choice.

Original languageEnglish
Title of host publication2021 International Conference on Circuits, Controls and Communications, CCUBE 2021
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781665402033
DOIs
Publication statusPublished - 2021
Event2021 International Conference on Circuits, Controls and Communications, CCUBE 2021 - Bangalore, India
Duration: 23-12-202124-12-2021

Publication series

Name2021 International Conference on Circuits, Controls and Communications, CCUBE 2021

Conference

Conference2021 International Conference on Circuits, Controls and Communications, CCUBE 2021
Country/TerritoryIndia
CityBangalore
Period23-12-2124-12-21

All Science Journal Classification (ASJC) codes

  • Control and Systems Engineering
  • Hardware and Architecture

Fingerprint

Dive into the research topics of 'Performance Analysis of the Pretrained EfficientDet for Real-time Object Detection on Raspberry Pi'. Together they form a unique fingerprint.

Cite this