Lightweight transformer image feature extraction network
PeerJ Computer Science · 31.01.2024 · 10 · e1755
DOI: 10.7717/peerj-cs.1755 ↗Kurz gefasst
Fachartikel in der Zeitschrift PeerJ Computer Science, online erschienen am 31.01.2024. Verfasst von Wenfeng Zheng und 4 weiteren Autor:innen. Beteiligte Einrichtung: School of Automation, University of Electronic Science and Technology of China, Chengdu, Sichuan, China. Die Arbeit wurde bisher 134-mal zitiert (Crossref) und nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Sichuan Science and Technology Program.
Abstract (Originalsprache)
In recent years, the image feature extraction method based on Transformer has become a research hotspot. However, when using Transformer for image feature extraction, the model’s complexity increases quadratically with the number of tokens entered. The quadratic complexity prevents vision transformer-based backbone networks from modelling high-resolution images and is computationally expensive. To address this issue, this study proposes two approaches to speed up Transformer models. Firstly, the self-attention mechanism’s quadratic complexity is reduced to linear, enhancing the model’s internal processing speed. Next, a parameter-less lightweight pruning method is introduced, which adaptively samples input images to filter out unimportant tokens, effectively reducing irrelevant input. Finally, these two methods are combined to create an efficient attention mechanism. Experimental results demonstrate that the combined methods can reduce the computation of the original Transformer model by 30%–50%, while the efficient attention mechanism achieves an impressive 60%–70% reduction in computation.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- School of Automation, University of Electronic Science and Technology of China, Chengdu, Sichuan, China.
- College of Resource and Environment Engineering, Guizhou University, Guiyang, Guizhou, China.
- Department of Geography and Anthropology, Louisiana State University, Baton Rouge, LA, United States of America.
Förderung
- Sichuan Science and Technology Program
Zitieren
APA
Zheng, W., Lu, S., Yang, Y., Yin, Z., & Yin, L. (2024). Lightweight transformer image feature extraction network. PeerJ Computer Science, 10, e1755. https://doi.org/10.7717/peerj-cs.1755
Vancouver
Zheng W, Lu S, Yang Y, Yin Z, Yin L. Lightweight transformer image feature extraction network. PeerJ Computer Science. 2024;10:e1755. doi:10.7717/peerj-cs.1755
BibTeX
@article{Zheng2024,
author = {Zheng, Wenfeng and Lu, Siyu and Yang, Youshuai and Yin, Zhengtong and Yin, Lirong},
title = {Lightweight transformer image feature extraction network},
journal = {PeerJ Computer Science},
year = {2024},
volume = {10},
pages = {e1755},
doi = {10.7717/peerj-cs.1755}
}
EUVIMED