Are Visual Place Recognition Models Recognizing Places or Conditions? Distractor-Augmented Evaluation and Condition Suppression

2026-08-07

AI Summary

Existing visual place recognition (VPR) models often conflate similarity in illumination, weather, or other transient conditions with true place identity, leading to degraded performance in the presence of distractors. To address this, this work proposes the Distractor-Augmented Recall (DAR) evaluation metric, which for the first time systematically quantifies a model’s sensitivity to condition-based interference. Furthermore, it integrates conditional suppression techniques—such as INLP and LEACE—to explicitly disentangle non-place-related condition information from feature descriptors. Experiments across 11 methods and 6 datasets demonstrate that performance rankings under DAR@1 significantly diverge from those under conventional Recall@1. Crucially, applying condition suppression consistently improves DAR@1 without compromising standard recall, confirming its efficacy in enhancing the robustness of place recognition systems.

Abstract

Long-term Visual Place Recognition (VPR) is typically evaluated by matching queries from one condition against a database from another. Crowdsourced map databases, however, may mix conditions and include images that resemble the query in condition but depict different places. In the presence of these distractors, a method may retrieve by condition similarity rather than place identity. We argue that this susceptibility arises because the discriminability of VPR methods allows them to encode information such as illumination, weather, and seasonal appearance in their descriptors. We therefore introduce Distractor-Augmented Recall (DAR) to isolate and quantify the effect of distractors, and propose condition suppression to remove condition information from VPR descriptors. Across eleven methods and six datasets, method rankings under DAR@1 differ from those under Recall@1 (R@1), while applying INLP and LEACE as condition suppression methods generally improves DAR@1 without reducing R@1. Thus, distractor robustness is distinct from standard retrieval performance and can be improved by suppressing condition information.

Problem

• Visual Place Recognition
• condition similarity
• distractors
• long-term VPR
• place identity

Innovation

• Visual Place Recognition
• Distractor-Augmented Evaluation
• Condition Suppression
• Feature Disentanglement
• Robust Retrieval