Spatial-Frequency Domain Aggregation for Visual Place Recognition


Chaoqun Wang (South China Normal University), Shaobo Min (Tencent)
The 35th British Machine Vision Conference
PDFPosterVideo (Right click to download)

Abstract

Visual place recognition (VPR) aims to identify an image by matching its global representation with those of reference images. Existing methods primarily aggregate features in the spatial domain, usually ignoring critical global or structural information embedded in the frequency domain. To address this limitation, we propose Dual-Domain Aggregation for Visual Place Recognition (DDA-VPR), which jointly integrates spatial and frequency domain features to enhance global representation aggregation. The core motivation is that the frequency domain inherently captures global and structural patterns, offering complementary and discriminative cues that are difficult to capture in the spatial domain. To bridge the domain gap between spatial and frequency features, a triple fusion strategy is introduced to facilitate cross-domain interaction and fuse spatial and frequency domain features into a unified and robust global representation. Additionally, to refine feature abstraction in both domains, a multi-scale contextual attention module is designed to leverage multi-scale information and preserve critical details during downsampling. Consequently, DDA-VPR generates more discriminative representations than methods that aggregate features solely in the spatial domain. Extensive experiments on challenging benchmarks demonstrate the superior performance of DDA-VPR, validating the effectiveness of dual-domain aggregation in VPR.

Citation

@inproceedings{Wang_2025_BMVC,
author    = {Chaoqun Wang and Shaobo Min},
title     = {Spatial-Frequency Domain Aggregation for Visual Place Recognition},
booktitle = {36th British Machine Vision Conference 2025, {BMVC} 2025, Sheffield, UK, November 24-27, 2025},
publisher = {BMVA},
year      = {2025},
url       = {https://bmva-archive.org.uk/bmvc/2025/assets/papers/Paper_147/paper.pdf}
}


Copyright © 2025 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection