Seqgeo Vl

SeqGeo-VL is a sequential cross-view geo-localization dataset pairing satellite imagery, street-view image sequences, GPS trajectories, and natural-language route descriptions. The dataset was created by author jiayulin and was last updated on July 15, 2026. This repository contains annotations and metadata in JSON format, while the images are redistributed from their original sources under separate access terms.

Use Cases

• Train cross-view image retrieval models based on paired satellite and street-view sequences.
• Develop sequence-to-sequence models for geo-localization based on GPS trajectory data.
• Build multimodal systems that align natural language route descriptions with visual geospatial data.
• Research spatio-temporal reasoning in AI by leveraging sequential image and trajectory data.

Strengths

• Integrates four distinct data modalities: satellite imagery, street-view sequences, GPS trajectories, and natural language.
• Focuses on sequential data, which likely captures temporal and spatial dynamics for geo-localization tasks.
• Last updated on 2026-07-15 17:37:11, indicating recent maintenance.

Limitations

• The repository contains only annotations and metadata; the actual satellite and street-view images are not included and must be obtained separately.
• Row count, column definitions, and file formats are unknown, limiting suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

huggingface

Collection Method

Annotations and metadata compiled by author jiayulin; images redistributed from original sources.

Freshness

Last updated 2026-07-15 17:37:11