Leveraging Selective Prediction for Reliable Image Geolocation

TL;DR

This paper introduces a method to assess whether images are suitable for geolocation, significantly improving the reliability and accuracy of image-based location estimation by selectively abstaining from uncertain images.

Contribution

The paper proposes two novel selection functions for assessing image localizability, enhancing the reliability of geolocation models by filtering out non-localizable images.

Findings

Improved city-scale geolocation accuracy from 27.8% to 70.5%.

Outperformed existing selective prediction baselines.

Enhanced reliability of geolocation models for real-world use.

Abstract

Reliable image geolocation is crucial for several applications, ranging from social media geo-tagging to fake news detection. State-of-the-art geolocation methods surpass human performance on the task of geolocation estimation from images. However, no method assesses the suitability of an image for this task, which results in unreliable and erroneous estimations for images containing no geolocation clues. In this paper, we define the task of image localizability, i.e. suitability of an image for geolocation, and propose a selective prediction methodology to address the task. In particular, we propose two novel selection functions that leverage the output probability distributions of geolocation models to infer localizability at different scales. Our selection functions are benchmarked against the most widely used selective prediction baselines, outperforming them in all cases. By abstaining from predicting non-localizable images, we improve geolocation accuracy from 27.8% to 70.5% at the city-scale, and thus make current geolocation models reliable for real-world applications.