Crowd Counting on Heavily Compressed Images with Curriculum Pre-Training

Arian Bakhtiarnia; Qi Zhang; Alexandros Iosifidis

doi:10.1109/SSCI52147.2023.10371805

Crowd Counting on Heavily Compressed Images with Curriculum Pre-Training

Arian Bakhtiarnia, Qi Zhang, Alexandros Iosifidis

DIGIT

Research output: Contribution to book/anthology/report/proceeding › Article in proceedings › Research › peer-review

Abstract

JPEG image compression algorithm is a widely used technique for image size reduction in edge and cloud computing settings. However, applying such lossy compression on images processed by deep neural networks can lead to significant accuracy degradation. Inspired by the curriculum learning paradigm, we propose a training approach called curriculum pre-training (CPT) for crowd counting on compressed images, which alleviates the drop in accuracy resulting from lossy compression. We verify the effectiveness of our approach by extensive experiments on three crowd counting datasets, two crowd counting DNN models and various levels of compression. The proposed training method is not overly sensitive to hyperparameters, and reduces the error, particularly for heavily compressed images, by up to 19.70%.

Original language	English
Title of host publication	2023 IEEE Symposium Series on Computational Intelligence (SSCI)
Number of pages	6
Publisher	IEEE
Publication date	2023
Pages	559-564
ISBN (Electronic)	978-1-6654-3065-4, 978-1-6654-3064-7
DOIs	https://doi.org/10.1109/SSCI52147.2023.10371805
Publication status	Published - 2023

Series	Proceedings (IEEE Symposium Series on Computational Intelligence)
ISSN	2770-0097

Keywords

Computer Vision
Crowd Counting
Smart City

Access to Document

10.1109/SSCI52147.2023.10371805

https://arxiv.org/pdf/2208.07075Licence: Unspecified

Cite this

@inproceedings{7b8c8718a451489c921b62bbeee08166,

title = "Crowd Counting on Heavily Compressed Images with Curriculum Pre-Training",

abstract = "JPEG image compression algorithm is a widely used technique for image size reduction in edge and cloud computing settings. However, applying such lossy compression on images processed by deep neural networks can lead to significant accuracy degradation. Inspired by the curriculum learning paradigm, we propose a training approach called curriculum pre-training (CPT) for crowd counting on compressed images, which alleviates the drop in accuracy resulting from lossy compression. We verify the effectiveness of our approach by extensive experiments on three crowd counting datasets, two crowd counting DNN models and various levels of compression. The proposed training method is not overly sensitive to hyperparameters, and reduces the error, particularly for heavily compressed images, by up to 19.70%.",

keywords = "Computer Vision, Crowd Counting, Smart City",

author = "Arian Bakhtiarnia and Qi Zhang and Alexandros Iosifidis",

year = "2023",

doi = "10.1109/SSCI52147.2023.10371805",

language = "English",

series = "Proceedings (IEEE Symposium Series on Computational Intelligence)",

publisher = "IEEE",

pages = "559--564",

booktitle = "2023 IEEE Symposium Series on Computational Intelligence (SSCI)",

}

Crowd Counting on Heavily Compressed Images with Curriculum Pre-Training. / Bakhtiarnia, Arian; Zhang, Qi; Iosifidis, Alexandros.
2023 IEEE Symposium Series on Computational Intelligence (SSCI). IEEE, 2023. p. 559-564 (Proceedings (IEEE Symposium Series on Computational Intelligence)).

Research output: Contribution to book/anthology/report/proceeding › Article in proceedings › Research › peer-review

TY - GEN

T1 - Crowd Counting on Heavily Compressed Images with Curriculum Pre-Training

AU - Bakhtiarnia, Arian

AU - Zhang, Qi

AU - Iosifidis, Alexandros

PY - 2023

Y1 - 2023

N2 - JPEG image compression algorithm is a widely used technique for image size reduction in edge and cloud computing settings. However, applying such lossy compression on images processed by deep neural networks can lead to significant accuracy degradation. Inspired by the curriculum learning paradigm, we propose a training approach called curriculum pre-training (CPT) for crowd counting on compressed images, which alleviates the drop in accuracy resulting from lossy compression. We verify the effectiveness of our approach by extensive experiments on three crowd counting datasets, two crowd counting DNN models and various levels of compression. The proposed training method is not overly sensitive to hyperparameters, and reduces the error, particularly for heavily compressed images, by up to 19.70%.

AB - JPEG image compression algorithm is a widely used technique for image size reduction in edge and cloud computing settings. However, applying such lossy compression on images processed by deep neural networks can lead to significant accuracy degradation. Inspired by the curriculum learning paradigm, we propose a training approach called curriculum pre-training (CPT) for crowd counting on compressed images, which alleviates the drop in accuracy resulting from lossy compression. We verify the effectiveness of our approach by extensive experiments on three crowd counting datasets, two crowd counting DNN models and various levels of compression. The proposed training method is not overly sensitive to hyperparameters, and reduces the error, particularly for heavily compressed images, by up to 19.70%.

KW - Computer Vision

KW - Crowd Counting

KW - Smart City

UR - http://www.scopus.com/inward/record.url?scp=85182937617&partnerID=8YFLogxK

U2 - 10.1109/SSCI52147.2023.10371805

DO - 10.1109/SSCI52147.2023.10371805

M3 - Article in proceedings

T3 - Proceedings (IEEE Symposium Series on Computational Intelligence)

SP - 559

EP - 564

BT - 2023 IEEE Symposium Series on Computational Intelligence (SSCI)

PB - IEEE

ER -

Crowd Counting on Heavily Compressed Images with Curriculum Pre-Training

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this