University-1652 is a benchmark dataset for drone-based geo-localization, cited in 50+ papers. It contains over 139,000 images across drone, satellite, street, and Google views, annotating 1,652 buildings from 72 universities. The dataset was published for ACM Multimedia 2020 and is hosted on Hugging Face by author layumi.
Use Cases
- Train models for drone-to-satellite image matching based on the multi-view structure.
- Develop algorithms for satellite-to-drone navigation using the provided query and gallery splits.
- Benchmark cross-view retrieval performance for academic research.
- Study building recognition and localization across different aerial and ground-level perspectives.
Strengths
- Annotates 1,652 distinct buildings, providing a substantial object base.
- Includes images from four distinct viewpoints: drone, satellite, street, and Google.
- Provides a defined train/test split with over 50,000 training images and tens of thousands of query/gallery images.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Data may reflect geographic bias inherent to its collection from 72 university campuses.
Provenance
- Source
- Hugging Face dataset uploaded by author 'layumi'.
- Collection Method
- Likely collected and annotated for academic research, as referenced in the ACM Multimedia 2020 paper.
- Time Range
- null
- Freshness
- Last updated 2025-10-24 14:44:35; freshness should be verified.
- Geography
- Buildings across 72 universities; specific locations are not detailed.