Loading [MathJax]/extensions/tex2jax.js

SparseAlign: A Fully Sparse Framework for Cooperative Object Detection

Publikation: Beitrag in FachzeitschriftKonferenzaufsatz in FachzeitschriftForschungPeer-Review

Autorschaft

Externe Organisationen

  • Technische Universität München (TUM)
  • Munich Center for Machine Learning (MCML)

Details

OriginalspracheEnglisch
FachzeitschriftCVPR
PublikationsstatusElektronisch veröffentlicht (E-Pub) - 17 März 2025

Abstract

Cooperative perception can increase the view field and decrease the occlusion of an ego vehicle, hence improving the perception performance and safety of autonomous driving. Despite the success of previous works on cooperative object detection, they mostly operate on dense Bird's Eye View (BEV) feature maps, which are computationally demanding and can hardly be extended to long-range detection problems. More efficient fully sparse frameworks are rarely explored. In this work, we design a fully sparse framework, SparseAlign, with three key features: an enhanced sparse 3D backbone, a query-based temporal context learning module, and a robust detection head specially tailored for sparse features. Extensive experimental results on both OPV2V and DairV2X datasets show that our framework, despite its sparsity, outperforms the state of the art with less communication bandwidth requirements. In addition, experiments on the OPV2Vt and DairV2Xt datasets for time-aligned cooperative object detection also show a significant performance gain compared to the baseline works.

Zitieren

SparseAlign: A Fully Sparse Framework for Cooperative Object Detection. / Yuan, Yunshuang; Xia, Yan; Cremers, Daniel et al.
in: CVPR, 17.03.2025.

Publikation: Beitrag in FachzeitschriftKonferenzaufsatz in FachzeitschriftForschungPeer-Review

Yuan Y, Xia Y, Cremers D, Sester M. SparseAlign: A Fully Sparse Framework for Cooperative Object Detection. CVPR. 2025 Mär 17. Epub 2025 Mär 17. doi: 10.48550/arXiv.2503.12982
Yuan, Yunshuang ; Xia, Yan ; Cremers, Daniel et al. / SparseAlign : A Fully Sparse Framework for Cooperative Object Detection. in: CVPR. 2025.
Download
@article{c10f90057d224336a895bd0f17273500,
title = "SparseAlign: A Fully Sparse Framework for Cooperative Object Detection",
abstract = "Cooperative perception can increase the view field and decrease the occlusion of an ego vehicle, hence improving the perception performance and safety of autonomous driving. Despite the success of previous works on cooperative object detection, they mostly operate on dense Bird's Eye View (BEV) feature maps, which are computationally demanding and can hardly be extended to long-range detection problems. More efficient fully sparse frameworks are rarely explored. In this work, we design a fully sparse framework, SparseAlign, with three key features: an enhanced sparse 3D backbone, a query-based temporal context learning module, and a robust detection head specially tailored for sparse features. Extensive experimental results on both OPV2V and DairV2X datasets show that our framework, despite its sparsity, outperforms the state of the art with less communication bandwidth requirements. In addition, experiments on the OPV2Vt and DairV2Xt datasets for time-aligned cooperative object detection also show a significant performance gain compared to the baseline works.",
keywords = "cs.CV",
author = "Yunshuang Yuan and Yan Xia and Daniel Cremers and Monika Sester",
note = "DBLP License: DBLP's bibliographic metadata records provided through http://dblp.org/ are distributed under a Creative Commons CC0 1.0 Universal Public Domain Dedication. Although the bibliographic metadata records are provided consistent with CC0 1.0 Dedication, the content described by the metadata records is not. Content may be subject to copyright, rights of privacy, rights of publicity and other restrictions.",
year = "2025",
month = mar,
day = "17",
doi = "10.48550/arXiv.2503.12982",
language = "English",

}

Download

TY - JOUR

T1 - SparseAlign

T2 - A Fully Sparse Framework for Cooperative Object Detection

AU - Yuan, Yunshuang

AU - Xia, Yan

AU - Cremers, Daniel

AU - Sester, Monika

N1 - DBLP License: DBLP's bibliographic metadata records provided through http://dblp.org/ are distributed under a Creative Commons CC0 1.0 Universal Public Domain Dedication. Although the bibliographic metadata records are provided consistent with CC0 1.0 Dedication, the content described by the metadata records is not. Content may be subject to copyright, rights of privacy, rights of publicity and other restrictions.

PY - 2025/3/17

Y1 - 2025/3/17

N2 - Cooperative perception can increase the view field and decrease the occlusion of an ego vehicle, hence improving the perception performance and safety of autonomous driving. Despite the success of previous works on cooperative object detection, they mostly operate on dense Bird's Eye View (BEV) feature maps, which are computationally demanding and can hardly be extended to long-range detection problems. More efficient fully sparse frameworks are rarely explored. In this work, we design a fully sparse framework, SparseAlign, with three key features: an enhanced sparse 3D backbone, a query-based temporal context learning module, and a robust detection head specially tailored for sparse features. Extensive experimental results on both OPV2V and DairV2X datasets show that our framework, despite its sparsity, outperforms the state of the art with less communication bandwidth requirements. In addition, experiments on the OPV2Vt and DairV2Xt datasets for time-aligned cooperative object detection also show a significant performance gain compared to the baseline works.

AB - Cooperative perception can increase the view field and decrease the occlusion of an ego vehicle, hence improving the perception performance and safety of autonomous driving. Despite the success of previous works on cooperative object detection, they mostly operate on dense Bird's Eye View (BEV) feature maps, which are computationally demanding and can hardly be extended to long-range detection problems. More efficient fully sparse frameworks are rarely explored. In this work, we design a fully sparse framework, SparseAlign, with three key features: an enhanced sparse 3D backbone, a query-based temporal context learning module, and a robust detection head specially tailored for sparse features. Extensive experimental results on both OPV2V and DairV2X datasets show that our framework, despite its sparsity, outperforms the state of the art with less communication bandwidth requirements. In addition, experiments on the OPV2Vt and DairV2Xt datasets for time-aligned cooperative object detection also show a significant performance gain compared to the baseline works.

KW - cs.CV

U2 - 10.48550/arXiv.2503.12982

DO - 10.48550/arXiv.2503.12982

M3 - Conference article

JO - CVPR

JF - CVPR

ER -

Von denselben Autoren