Density-Aware DETR With Dynamic Query for End-to-End Tiny Object Detection

End-to-end DEtection TRansformer (DETRs) are leading a new trend in various object detection tasks. However, when it comes to the ubiquitous tiny objects in aerial imagery, the potential of DETRs still remains under-explored. In this work, we observe that the expansive field of view of remote sensin...

Full description

Saved in:
Bibliographic Details
Main Authors: Xianhang Ye, Chang Xu, Haoran Zhu, Fang Xu, Haijian Zhang, Wen Yang
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/11007261/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849472105032187904
author Xianhang Ye
Chang Xu
Haoran Zhu
Fang Xu
Haijian Zhang
Wen Yang
author_facet Xianhang Ye
Chang Xu
Haoran Zhu
Fang Xu
Haijian Zhang
Wen Yang
author_sort Xianhang Ye
collection DOAJ
description End-to-end DEtection TRansformer (DETRs) are leading a new trend in various object detection tasks. However, when it comes to the ubiquitous tiny objects in aerial imagery, the potential of DETRs still remains under-explored. In this work, we observe that the expansive field of view of remote sensing images often results in a limited pixel representation of tiny objects coupled with a substantial variance in the number of instances across images. The significantly varied tiny object number per image conflicts with DETRs' fixed set of object queries. A large number of queries are necessary to ensure high recall in dense scenarios, while sparse scenarios benefit from fewer, more distinct queries. To tackle this issue, we propose a Density-aware DETR with Dynamic Query (D3Q). D3Q adaptively determines the optimal number of object queries for each image by explicitly estimating its object density. This dynamic query mechanism enables efficient and accurate tiny object detection under both dense and sparse object distributions. In addition, we introduce a refined box loss designed for tiny object detection that further stabilizes training. Through these strategies, D3Q effectively adapts to both dense and sparse scenarios, overcoming the limitations of fixed query in DETR. Extensive experiments on challenging tiny object detection benchmarks demonstrate the superior performance of D3Q compared to state-of-the-art methods. Particularly, when integrated with DINO, D3Q achieves an impressive 32.1% mAP on the AI-TOD-v2 dataset, setting a new state-of-the-art performance.
format Article
id doaj-art-6f13a2d1e93d40ffac16134d94b2ac4f
institution Kabale University
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-6f13a2d1e93d40ffac16134d94b2ac4f2025-08-20T03:24:37ZengIEEEIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing1939-14042151-15352025-01-0118135541356910.1109/JSTARS.2025.357181411007261Density-Aware DETR With Dynamic Query for End-to-End Tiny Object DetectionXianhang Ye0https://orcid.org/0009-0003-3334-4686Chang Xu1https://orcid.org/0000-0002-3078-0496Haoran Zhu2https://orcid.org/0009-0003-0153-1305Fang Xu3https://orcid.org/0000-0003-4260-7911Haijian Zhang4https://orcid.org/0000-0001-8314-6563Wen Yang5https://orcid.org/0000-0002-3263-8768School of Electronic Information, Wuhan University, Wuhan, ChinaSchool of Electronic Information, Wuhan University, Wuhan, ChinaSchool of Electronic Information, Wuhan University, Wuhan, ChinaSchool of Artificial Intelligence, Wuhan University, Wuhan, ChinaSchool of Electronic Information, Wuhan University, Wuhan, ChinaSchool of Electronic Information, Wuhan University, Wuhan, ChinaEnd-to-end DEtection TRansformer (DETRs) are leading a new trend in various object detection tasks. However, when it comes to the ubiquitous tiny objects in aerial imagery, the potential of DETRs still remains under-explored. In this work, we observe that the expansive field of view of remote sensing images often results in a limited pixel representation of tiny objects coupled with a substantial variance in the number of instances across images. The significantly varied tiny object number per image conflicts with DETRs' fixed set of object queries. A large number of queries are necessary to ensure high recall in dense scenarios, while sparse scenarios benefit from fewer, more distinct queries. To tackle this issue, we propose a Density-aware DETR with Dynamic Query (D3Q). D3Q adaptively determines the optimal number of object queries for each image by explicitly estimating its object density. This dynamic query mechanism enables efficient and accurate tiny object detection under both dense and sparse object distributions. In addition, we introduce a refined box loss designed for tiny object detection that further stabilizes training. Through these strategies, D3Q effectively adapts to both dense and sparse scenarios, overcoming the limitations of fixed query in DETR. Extensive experiments on challenging tiny object detection benchmarks demonstrate the superior performance of D3Q compared to state-of-the-art methods. Particularly, when integrated with DINO, D3Q achieves an impressive 32.1% mAP on the AI-TOD-v2 dataset, setting a new state-of-the-art performance.https://ieeexplore.ieee.org/document/11007261/Density estimationdetection transformer (DETR)dynamic query (D3Q)label assignmenttiny object detection
spellingShingle Xianhang Ye
Chang Xu
Haoran Zhu
Fang Xu
Haijian Zhang
Wen Yang
Density-Aware DETR With Dynamic Query for End-to-End Tiny Object Detection
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Density estimation
detection transformer (DETR)
dynamic query (D3Q)
label assignment
tiny object detection
title Density-Aware DETR With Dynamic Query for End-to-End Tiny Object Detection
title_full Density-Aware DETR With Dynamic Query for End-to-End Tiny Object Detection
title_fullStr Density-Aware DETR With Dynamic Query for End-to-End Tiny Object Detection
title_full_unstemmed Density-Aware DETR With Dynamic Query for End-to-End Tiny Object Detection
title_short Density-Aware DETR With Dynamic Query for End-to-End Tiny Object Detection
title_sort density aware detr with dynamic query for end to end tiny object detection
topic Density estimation
detection transformer (DETR)
dynamic query (D3Q)
label assignment
tiny object detection
url https://ieeexplore.ieee.org/document/11007261/
work_keys_str_mv AT xianhangye densityawaredetrwithdynamicqueryforendtoendtinyobjectdetection
AT changxu densityawaredetrwithdynamicqueryforendtoendtinyobjectdetection
AT haoranzhu densityawaredetrwithdynamicqueryforendtoendtinyobjectdetection
AT fangxu densityawaredetrwithdynamicqueryforendtoendtinyobjectdetection
AT haijianzhang densityawaredetrwithdynamicqueryforendtoendtinyobjectdetection
AT wenyang densityawaredetrwithdynamicqueryforendtoendtinyobjectdetection