SeqGeo-VL: Sequential Cross-View Geo-Localization Dataset

A multimodal dataset pairing satellite imagery, street-view image sequences, GPS trajectories, and natural-language route descriptions for cross-view geo-localization tasks. The dataset was created by MVRL and its metadata was last updated on July 15, 2026. The repository contains annotations and metadata in JSON format, while the images themselves are redistributed from their original sources.

Use Cases

• Train models for cross-view image retrieval based on paired satellite and street-view sequences.
• Develop sequential geo-localization algorithms using GPS trajectory data mentioned in the description.
• Build models for generating or interpreting natural-language route descriptions from visual data.
• Benchmark multimodal fusion techniques for tasks linking imagery, trajectories, and text.

Strengths

• Dataset integrates four distinct data modalities: satellite imagery, street-view sequences, GPS trajectories, and text descriptions.
• Metadata was last updated on July 15, 2026, suggesting recent maintenance.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

MVRL

Collection Method

Annotations/metadata compiled from multiple original image sources.

Freshness

Last updated 2026-07-15 17:37:11; freshness should be verified.