Just Zoom In: Cross-View Geo-Localization Dataset for Washington, D.C.

A dataset for cross-view geo-localization, pairing ground-level street-view images with overhead satellite or aerial imagery. It was created by pcvlab and last updated on 2026-05-17. The dataset focuses on Washington, D.C., and uses a multi-scale tile hierarchy for realistic model evaluation.

Use Cases

• Train models to match a street-view image to its corresponding overhead tile based on the described image pairs.
• Evaluate cross-view geo-localization algorithms using the described multi-scale satellite/aerial tile hierarchy.
• Benchmark performance on a realistic, limited-field-of-view street-view dataset as described.
• Develop autoregressive zooming techniques for geo-localization as referenced in the accompanying paper.

Strengths

• Pairs crowd-sourced street-view images with geo-referenced overhead imagery, as described.
• Focuses on a specific, well-defined geographic area: Washington, D.C.
• Designed for realistic evaluation with a multi-scale tile hierarchy.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and dataset size are unknown, which may limit suitability assessment.
• Data may reflect geographic bias inherent to its focus on a single city.

Provenance

Source

pcvlab

Collection Method

Pairs crowd-sourced street-view images with overhead imagery.

Freshness

Last updated 2026-05-17 12:22:03; freshness should be verified.

Geography

Washington, D.C.