University-1652: Multi-View Geo-Localization of 1,652 Buildings

University-1652 is a multi-view multi-source benchmark for drone-based geo-localization released by author layumi for ACM Multimedia 2020. It contains annotations for 1,652 buildings across 72 universities globally, providing a framework for matching drone-view queries against satellite-view galleries. The data supports cross-view image retrieval tasks using imagery from drones, satellites, and ground-level sources.

Use Cases

• Cross-view image retrieval matching drone-view queries to satellite-view galleries
• Geo-localization of specific buildings using multi-source visual features
• Training deep learning models with GeM-pooling for remote sensing applications

Strengths

• Annotates 1,652 unique buildings
• Includes multi-source imagery from drone, satellite, and ground views
• Global geographic scope covering 72 universities
• MIT licensed for open academic and commercial use

Limitations

• Geographically restricted to university campus environments
• Potential domain gap between high-altitude satellite sensors and low-altitude drone cameras
• Lacks tabular metadata for non-visual building attributes

Provenance

Source

ACM Multimedia 2020 publication by layumi

Collection Method

Annotated multi-source imagery collection from drone, satellite, and ground perspectives

Freshness

Last updated February 2026 according to repository metadata.

Geography

Global (72 universities)