VPR-AttLLM enhances existing visual place recognition models by using LLMs to focus on location-informative image regions, delivering 1-3% recall gains and up to 8% on real flood imagery from San Francisco and Hong Kong.
Correct" denotesattentionmapsgeneratedwithaccurategeographiccontext(SFforSanFranciscoqueries,HKforHongKong queries)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Enhancing Geo-localization for Crowdsourced Flood Imagery via LLM-Guided Attention
VPR-AttLLM enhances existing visual place recognition models by using LLMs to focus on location-informative image regions, delivering 1-3% recall gains and up to 8% on real flood imagery from San Francisco and Hong Kong.