Selectivity Drives Efficiency: Dataset Pruning for Visual Place Recognition

2026-07-16

AI Summary

This work addresses the high storage and training costs of large-scale visual place recognition (VPR) datasets, a challenge exacerbated by existing pruning methods that overlook the relational dependencies inherent in VPR. The study introduces the first location-level data pruning framework, elevating the pruning unit from individual images to entire places. It jointly evaluates the training utility of each location through two complementary metrics: intra-place diversity (IPD) and inter-place similarity (IPS), aligning with the relational supervision nature of VPR. Using the IPD-IPS scoring mechanism, the authors construct highly efficient training subsets that significantly outperform current approaches. On mainstream models such as NetVLAD, their method achieves 94.5% and 97.0% R@1 accuracy on MSLS-val and Nordland, respectively, after compressing the dataset to less than one-third the size of GSV-Cities.

Abstract

Recent visual place recognition (VPR) studies have increasingly relied on large-scale datasets to train more robust and discriminative models. Although this trend significantly improves recognition performance, it also introduces substantial storage and training costs, especially when new architectures or training strategies need to be repeatedly developed and evaluated. Dataset pruning (DP) provides a promising way to improve data efficiency by retaining only informative training data. However, conventional DP methods mainly follow the sample-wise classification paradigm, which overlooks the relation-dependent training nature of VPR, where supervision is typically formed by image pairs rather than independent images. To address this issue, we propose a place-wise dataset pruning framework tailored for VPR. Instead of pruning individual images, our method treats each place as the basic pruning unit and introduces two complementary novel metrics, i.e., intra-place diversity (IPD) and inter-place similarity (IPS), to evaluate the training value of each place. By jointly considering these two metrics, our method ranks all places and constructs a compact yet informative coreset, thereby allowing the pruned dataset to still support the training of robust and discriminative VPR models. Extensive experiments demonstrate that our method consistently outperforms state-of-the-art DP baselines under different pruning ratios while reducing selection and training costs. Moreover, by pruning a merged dataset roughly 3.5$\times$ the size of GSV-Cities to a comparable scale, our coreset maintains highly competitive performance, achieving 94.5\% R@1 on MSLS-val and 97.0\% R@1 on Nordland with only NetVLAD. Codes will be made publicly available.

Problem

• visual place recognition
• dataset pruning
• relation-dependent training
• data efficiency
• place-wise selection

Innovation

• dataset pruning
• visual place recognition
• intra-place diversity
• inter-place similarity
• coreset