G^3: Geolocation via Guidebook Grounding

TL;DR

This paper introduces a novel geolocation method that leverages human-written guidebooks to improve image location prediction, outperforming existing image-only approaches by over 5% in accuracy.

Contribution

It proposes the task of guidebook-grounded geolocation, utilizing textual clues from guidebooks to enhance visual geolocation accuracy, and introduces a new dataset and method for this purpose.

Findings

Outperforms state-of-the-art image-only geolocation methods by over 5% in Top-1 accuracy.

Supervising attention with country-level pseudo labels improves performance.

Uses a dataset of StreetView images and associated guidebooks for training and evaluation.

Abstract

We demonstrate how language can improve geolocation: the task of predicting the location where an image was taken. Here we study explicit knowledge from human-written guidebooks that describe the salient and class-discriminative visual features humans use for geolocation. We propose the task of Geolocation via Guidebook Grounding that uses a dataset of StreetView images from a diverse set of locations and an associated textual guidebook for GeoGuessr, a popular interactive geolocation game. Our approach predicts a country for each image by attending over the clues automatically extracted from the guidebook. Supervising attention with country-level pseudo labels achieves the best performance. Our approach substantially outperforms a state-of-the-art image-only geolocation method, with an improvement of over 5% in Top-1 accuracy. Our dataset and code can be found at https://github.com/g-luo/geolocation_via_guidebook_grounding.