To Match or Not to Match: Revisiting Image Matching for Reliable Visual Place Recognition

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Sferrazza, Davide, Berton, Gabriele, Trivigno, Gabriele, Masone, Carlo
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912340112310272
author Sferrazza, Davide
Berton, Gabriele
Trivigno, Gabriele
Masone, Carlo
author_facet Sferrazza, Davide
Berton, Gabriele
Trivigno, Gabriele
Masone, Carlo
contents Visual Place Recognition (VPR) is a critical task in computer vision, traditionally enhanced by re-ranking retrieval results with image matching. However, recent advancements in VPR methods have significantly improved performance, challenging the necessity of re-ranking. In this work, we show that modern retrieval systems often reach a point where re-ranking can degrade results, as current VPR datasets are largely saturated. We propose using image matching as a verification step to assess retrieval confidence, demonstrating that inlier counts can reliably predict when re-ranking is beneficial. Our findings shift the paradigm of retrieval pipelines, offering insights for more robust and adaptive VPR systems. The code is available at https://github.com/FarInHeight/To-Match-or-Not-to-Match.
format Preprint
id arxiv_https___arxiv_org_abs_2504_06116
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle To Match or Not to Match: Revisiting Image Matching for Reliable Visual Place Recognition
Sferrazza, Davide
Berton, Gabriele
Trivigno, Gabriele
Masone, Carlo
Computer Vision and Pattern Recognition
Visual Place Recognition (VPR) is a critical task in computer vision, traditionally enhanced by re-ranking retrieval results with image matching. However, recent advancements in VPR methods have significantly improved performance, challenging the necessity of re-ranking. In this work, we show that modern retrieval systems often reach a point where re-ranking can degrade results, as current VPR datasets are largely saturated. We propose using image matching as a verification step to assess retrieval confidence, demonstrating that inlier counts can reliably predict when re-ranking is beneficial. Our findings shift the paradigm of retrieval pipelines, offering insights for more robust and adaptive VPR systems. The code is available at https://github.com/FarInHeight/To-Match-or-Not-to-Match.
title To Match or Not to Match: Revisiting Image Matching for Reliable Visual Place Recognition
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2504.06116