Ultra-low power real-ime object detection based on quantized CNNs

With the recent proliferation of deep learning-based solutions to object detection, the state-of-the-art accuracy has been increasing far beyond what was achievable using traditional methods. However, the hardware requirements for running these models in real-time are high, so they are expensive to...

Full description

Saved in:
Bibliographic Details
Main Author: Chew, Jing Wei
Other Authors: Weichen Liu
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2021
Subjects:
Online Access:https://hdl.handle.net/10356/148048
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
id sg-ntu-dr.10356-148048
record_format dspace
spelling sg-ntu-dr.10356-1480482021-04-22T07:01:41Z Ultra-low power real-ime object detection based on quantized CNNs Chew, Jing Wei Weichen Liu School of Computer Science and Engineering liu@ntu.edu.sg Engineering::Computer science and engineering With the recent proliferation of deep learning-based solutions to object detection, the state-of-the-art accuracy has been increasing far beyond what was achievable using traditional methods. However, the hardware requirements for running these models in real-time are high, so they are expensive to deploy on the edge. Furthermore, due to their large model size, their memory footprint is unnecessarily high, and this also leads to excessive power consumption which makes them unfeasible for deployment on resource-constrained environments with no constant power source. Therefore, this project proposes the use of the most extreme network quantization possible, i.e. binarization, to make a YOLO-based object detection model deployable on the edge, while attaining reasonable accuracy. Using this approach, the proposed model can run at 37.7 FPS on an NVIDIA Jetson Nano with a peak memory footprint of 17.1 MB, while attaining a reasonable mAP@0.50 Intersection over Union (IoU) of 0.37 on the Pascal Visual Object Classes (VOC) dataset. Furthermore, these figures signify a speedup of 21.8x and a memory usage reduction by a factor of 15.3x compared to a similar YOLOv2 full-precision model architecture. Since computation was completely performed on the CPU, the use of TensorRT delegates or any other embedded hardware accelerator can allow for larger models with higher accuracies to be deployed in future works. The full project is open-sourced and can be found in https://github.com/tehtea/QuickYOLO. Bachelor of Engineering (Computer Science) 2021-04-22T07:01:41Z 2021-04-22T07:01:41Z 2021 Final Year Project (FYP) Chew, J. W. (2021). Ultra-low power real-ime object detection based on quantized CNNs. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/148048 https://hdl.handle.net/10356/148048 en application/pdf Nanyang Technological University
institution Nanyang Technological University
building NTU Library
continent Asia
country Singapore
Singapore
content_provider NTU Library
collection DR-NTU
language English
topic Engineering::Computer science and engineering
spellingShingle Engineering::Computer science and engineering
Chew, Jing Wei
Ultra-low power real-ime object detection based on quantized CNNs
description With the recent proliferation of deep learning-based solutions to object detection, the state-of-the-art accuracy has been increasing far beyond what was achievable using traditional methods. However, the hardware requirements for running these models in real-time are high, so they are expensive to deploy on the edge. Furthermore, due to their large model size, their memory footprint is unnecessarily high, and this also leads to excessive power consumption which makes them unfeasible for deployment on resource-constrained environments with no constant power source. Therefore, this project proposes the use of the most extreme network quantization possible, i.e. binarization, to make a YOLO-based object detection model deployable on the edge, while attaining reasonable accuracy. Using this approach, the proposed model can run at 37.7 FPS on an NVIDIA Jetson Nano with a peak memory footprint of 17.1 MB, while attaining a reasonable mAP@0.50 Intersection over Union (IoU) of 0.37 on the Pascal Visual Object Classes (VOC) dataset. Furthermore, these figures signify a speedup of 21.8x and a memory usage reduction by a factor of 15.3x compared to a similar YOLOv2 full-precision model architecture. Since computation was completely performed on the CPU, the use of TensorRT delegates or any other embedded hardware accelerator can allow for larger models with higher accuracies to be deployed in future works. The full project is open-sourced and can be found in https://github.com/tehtea/QuickYOLO.
author2 Weichen Liu
author_facet Weichen Liu
Chew, Jing Wei
format Final Year Project
author Chew, Jing Wei
author_sort Chew, Jing Wei
title Ultra-low power real-ime object detection based on quantized CNNs
title_short Ultra-low power real-ime object detection based on quantized CNNs
title_full Ultra-low power real-ime object detection based on quantized CNNs
title_fullStr Ultra-low power real-ime object detection based on quantized CNNs
title_full_unstemmed Ultra-low power real-ime object detection based on quantized CNNs
title_sort ultra-low power real-ime object detection based on quantized cnns
publisher Nanyang Technological University
publishDate 2021
url https://hdl.handle.net/10356/148048
_version_ 1698713740244942848