Geolocated Imagery Dataset (Scotland)

Key Use Cases

One of the primary defense applications for this dataset is to assist in locating hostages or other persons of interest from visual media such as photos or videos sent by captors. In such critical cases, intelligence analysts may only have limited visual data, such as one or two images, and may lack accurate metadata regarding the time or location of the capture. This dataset aids in the creation of machine learning algorithms capable of analyzing the content within the images—such as landscapes, buildings, or natural features—to predict the location.

Dataset Composition and Structure
The dataset consists of a variety of images representing both urban and rural environments, providing diverse visual cues such as buildings, streets, landscapes, and natural elements (e.g., forests, hills). Images were captured using a mobile phone, resulting in mixed quality, a deliberate decision to simulate real-world conditions where image quality may vary. These variations in quality and orientation challenge algorithm development, ensuring robustness across different scenarios.

Geolocation Accuracy

Each image in this dataset is accompanied by geolocation data, which has undergone manual inspection and corrections to mitigate inaccuracies caused by mobile GPS lag. The dataset boasts an accuracy within 100 meters for all images, with most achieving significantly higher accuracy. This ensures that the dataset is suitable for high-precision geolocation tasks, providing a reliable foundation for model training and evaluation.

Privacy and Data Modifications
To safeguard privacy, all identifiable faces and vehicle license plates in the images have been blurred. Apart from these modifications, the images have been kept in their original form to preserve the authenticity and realism of the dataset. The images span a range of angles, orientations, and times of day, making them ideal for training models that must function under variable conditions.

Data Quality and Variety

The dataset emphasizes realistic image conditions, offering a diverse range of visual content:

• Urban and rural settings.
• Varied lighting and weather conditions.
• Different image orientations and angles to simulate real-world data variability.