Selectivity Drives Efficiency: Dataset Pruning for Visual Place Recognition

2026-07-16 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

A new place-wise dataset pruning framework addresses the high storage and training costs associated with large-scale visual place recognition (VPR) datasets. Traditional dataset pruning methods, which are sample-wise, overlook VPR's inherent relation-dependent training nature, where supervision relies on image pairs. This novel approach treats each "place" as the fundamental pruning unit, introducing two complementary metrics: intra-place diversity (IPD) and inter-place similarity (IPS). By jointly evaluating these metrics, the framework ranks all places to create a compact yet informative coreset, enabling the training of robust and discriminative VPR models. Experiments show it consistently outperforms state-of-the-art baselines, reducing selection and training costs. Notably, pruning a dataset roughly 3.5x the size of GSV-Cities to a comparable scale achieved 94.5% R@1 on MSLS-val and 97.0% R@1 on Nordland using NetVLAD.

Key takeaway

For Computer Vision Engineers developing or evaluating visual place recognition (VPR) models, this place-wise dataset pruning framework offers a critical advantage. You can significantly reduce storage and training costs by curating a compact, informative coreset using intra-place diversity and inter-place similarity metrics. This approach maintains high recognition performance, as demonstrated by 94.5% R@1 on MSLS-val, allowing for more efficient model development and iteration. Consider adopting this method to optimize your VPR training pipelines.

Key insights

Place-wise dataset pruning, using intra-place diversity and inter-place similarity, significantly enhances visual place recognition data efficiency.

Principles

• VPR training relies on image pair relations.
• Informative data retention boosts efficiency.
• Place-wise pruning suits VPR better.

Method

The method treats each place as the basic pruning unit, evaluating its training value via intra-place diversity (IPD) and inter-place similarity (IPS). Places are ranked to construct a compact coreset.

In practice

• Reduce VPR dataset storage costs.
• Lower VPR model training expenses.
• Curate VPR datasets using IPD/IPS.

Topics

• Visual Place Recognition
• Dataset Pruning
• Data Efficiency
• Computer Vision
• Coreset Selection
• Intra-place Diversity

Best for: Research Scientist, AI Scientist, Computer Vision Engineer

Related on AIssential

• Automatic Map Density Selection for Locally-Performant Visual Place Recognition — A dynamic VPR mapping approach automatically selects optimal map density to meet user-defined local recall and achievement rate targets.
• Selecting Samples on Graphs: A Unified Dataset Pruning Framework for Lossless Training Acceleration — A unified graph-based framework prunes datasets by modeling samples as nodes and relations as edges, solving a Maximum Weight Clique Problem.
• Beyond Attention Scores: SVD-Based Vision Token Pruning for Efficient Vision-Language Models — SVD-Prune uses global variance and leverage scores for training-free VLM token pruning, outperforming attention-based methods.
• FlatVPR: Plug-and-play Geo-linear Residual Adapter for Geometric Rectification of Foundation Model Feature Manifolds — FlatVPR flattens foundation model feature manifolds for accurate VPR with sparse anchors via a learnable residual adapter and geometric loss.
• Quantile Transfer for Reliable Operating Point Selection in Visual Place Recognition — Quantile transfer robustly adapts VPR matching thresholds to maximize recall under user-defined precision, outperforming static methods.
• VisPCO: Visual Token Pruning Configuration Optimization via Budget-Aware Pareto-Frontier Learning for Vision-Language Models — VisPCO optimizes visual token pruning in VLMs by formulating it as a Pareto configuration problem for optimal efficiency.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.