Multi-scale Vertical Cross-layer Feature Aggregation and Attention Fusion Network for Object Detection

Wenting Gao, Xiaojuan Li, Yu Han, Yue Liu*

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

2 Citations (Scopus)

Abstract

Scale imbalance is one of the primary limitations for object detection. To tackle such a problem, existing methods such as FPN usually integrate the features at different scales, which suffers from the inconsistence of different high-level and low-level features due to the straightforward combination. In this paper, we propose a multi-scale vertical cross-layer feature aggregation and attention fusion network which not only has bottom-up and top-down pathways with lateral connections, but also adds cross-layer paths in the vertical direction. The proposed method can boost information flow and shorten the information path between high-level and low-level features. An attention fusion module is also introduced to obtain the internal correlation between local, global and contextual information of other feature layers. In order to optimize the anchor configurations, a differential evolution algorithm is employed to reconfigure the ratios and scales of anchors. Experimental results show that the proposed method achieves superior detection performance on the public dataset PASCAL VOC.

Original languageEnglish
Title of host publicationArtificial Neural Networks and Machine Learning - ICANN 2022 - 31st International Conference on Artificial Neural Networks, Proceedings
EditorsElias Pimenidis, Mehmet Aydin, Plamen Angelov, Chrisina Jayne, Antonios Papaleonidas
PublisherSpringer Science and Business Media Deutschland GmbH
Pages139-150
Number of pages12
ISBN (Print)9783031159367
DOIs
Publication statusPublished - 2022
Event31st International Conference on Artificial Neural Networks, ICANN 2022 - Bristol, United Kingdom
Duration: 6 Sept 20229 Sept 2022

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13532 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference31st International Conference on Artificial Neural Networks, ICANN 2022
Country/TerritoryUnited Kingdom
CityBristol
Period6/09/229/09/22

Keywords

  • Attention mechanism
  • Deep learning
  • Object detection

Fingerprint

Dive into the research topics of 'Multi-scale Vertical Cross-layer Feature Aggregation and Attention Fusion Network for Object Detection'. Together they form a unique fingerprint.

Cite this