Efficient Side-Tuning for Remote Sensing: A Low-Memory Fine-Tuning Framework

Fine-tuning pretrained models for remote sensing tasks often demands substantial computational resources. To reduce memory requirements and training costs, this article proposes a low-memory fine-tuning framework, called efficient side-tuning (EST), for remote sensing downstream tasks. EST attaches...

Full description

Saved in:
Bibliographic Details
Main Authors: Haichen Yu, Wenxin Yin, Hanbo Bi, Chongyang Li, Yingchao Feng, Wenhui Diao, Xian Sun
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10974700/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850138186389389312
author Haichen Yu
Wenxin Yin
Hanbo Bi
Chongyang Li
Yingchao Feng
Wenhui Diao
Xian Sun
author_facet Haichen Yu
Wenxin Yin
Hanbo Bi
Chongyang Li
Yingchao Feng
Wenhui Diao
Xian Sun
author_sort Haichen Yu
collection DOAJ
description Fine-tuning pretrained models for remote sensing tasks often demands substantial computational resources. To reduce memory requirements and training costs, this article proposes a low-memory fine-tuning framework, called efficient side-tuning (EST), for remote sensing downstream tasks. EST attaches a parallel network to the backbone of the model, and only fine-tunes the parameters of the parallel network during the training phase. The proposed EST Block is the main component of the parallel network, which uses the multichannel adapter fusion module, gate layer and depthwise convolution to achieve feature selection and enhancement effects. In the evaluation, on six remote sensing datasets including object detection and semantic segmentation, EST achieved SOTA performance results using only less than 40<inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> of the memory expenditure of full fine-tuning, which is better than all current parameter efficient fine-tuning methods. In addition, experiments on backbones of various sizes and classes show that the generalizability of EST is also reliable. EST thus offers a highly efficient and effective approach for efficient transfer learning in remote sensing, unlocking new possibilities for advanced remote sensing applications.
format Article
id doaj-art-e2995fe9b8e24264870c148dd5a56f65
institution OA Journals
issn 1939-1404
2151-1535
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
spelling doaj-art-e2995fe9b8e24264870c148dd5a56f652025-08-20T02:30:38ZengIEEEIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing1939-14042151-15352025-01-0118119081192510.1109/JSTARS.2025.356364110974700Efficient Side-Tuning for Remote Sensing: A Low-Memory Fine-Tuning FrameworkHaichen Yu0https://orcid.org/0009-0001-8123-417XWenxin Yin1https://orcid.org/0000-0003-0157-7947Hanbo Bi2https://orcid.org/0009-0001-4209-5461Chongyang Li3https://orcid.org/0009-0003-8234-4420Yingchao Feng4https://orcid.org/0000-0003-4017-8885Wenhui Diao5https://orcid.org/0000-0002-3931-3974Xian Sun6https://orcid.org/0000-0002-0038-9816Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing, ChinaAerospace Information Research Institute, Chinese Academy of Sciences, Beijing, ChinaAerospace Information Research Institute, Chinese Academy of Sciences, Beijing, ChinaAerospace Information Research Institute, Chinese Academy of Sciences, Beijing, ChinaAerospace Information Research Institute, Chinese Academy of Sciences, Beijing, ChinaAerospace Information Research Institute, Chinese Academy of Sciences, Beijing, ChinaAerospace Information Research Institute, Chinese Academy of Sciences, Beijing, ChinaFine-tuning pretrained models for remote sensing tasks often demands substantial computational resources. To reduce memory requirements and training costs, this article proposes a low-memory fine-tuning framework, called efficient side-tuning (EST), for remote sensing downstream tasks. EST attaches a parallel network to the backbone of the model, and only fine-tunes the parameters of the parallel network during the training phase. The proposed EST Block is the main component of the parallel network, which uses the multichannel adapter fusion module, gate layer and depthwise convolution to achieve feature selection and enhancement effects. In the evaluation, on six remote sensing datasets including object detection and semantic segmentation, EST achieved SOTA performance results using only less than 40<inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> of the memory expenditure of full fine-tuning, which is better than all current parameter efficient fine-tuning methods. In addition, experiments on backbones of various sizes and classes show that the generalizability of EST is also reliable. EST thus offers a highly efficient and effective approach for efficient transfer learning in remote sensing, unlocking new possibilities for advanced remote sensing applications.https://ieeexplore.ieee.org/document/10974700/Fine-tuningmemory efficienctobject detectionremote sensingsemantic segmentation
spellingShingle Haichen Yu
Wenxin Yin
Hanbo Bi
Chongyang Li
Yingchao Feng
Wenhui Diao
Xian Sun
Efficient Side-Tuning for Remote Sensing: A Low-Memory Fine-Tuning Framework
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Fine-tuning
memory efficienct
object detection
remote sensing
semantic segmentation
title Efficient Side-Tuning for Remote Sensing: A Low-Memory Fine-Tuning Framework
title_full Efficient Side-Tuning for Remote Sensing: A Low-Memory Fine-Tuning Framework
title_fullStr Efficient Side-Tuning for Remote Sensing: A Low-Memory Fine-Tuning Framework
title_full_unstemmed Efficient Side-Tuning for Remote Sensing: A Low-Memory Fine-Tuning Framework
title_short Efficient Side-Tuning for Remote Sensing: A Low-Memory Fine-Tuning Framework
title_sort efficient side tuning for remote sensing a low memory fine tuning framework
topic Fine-tuning
memory efficienct
object detection
remote sensing
semantic segmentation
url https://ieeexplore.ieee.org/document/10974700/
work_keys_str_mv AT haichenyu efficientsidetuningforremotesensingalowmemoryfinetuningframework
AT wenxinyin efficientsidetuningforremotesensingalowmemoryfinetuningframework
AT hanbobi efficientsidetuningforremotesensingalowmemoryfinetuningframework
AT chongyangli efficientsidetuningforremotesensingalowmemoryfinetuningframework
AT yingchaofeng efficientsidetuningforremotesensingalowmemoryfinetuningframework
AT wenhuidiao efficientsidetuningforremotesensingalowmemoryfinetuningframework
AT xiansun efficientsidetuningforremotesensingalowmemoryfinetuningframework