StreetVision-10K: Street-Level Photo Geolocation Dataset

StreetVision-10K is a dataset for training and evaluating geolocation models, created by author 'mishl'. Each sample includes a system prompt, a user message with a street-level photo, and an assistant response with ground-truth coordinates. The dataset was last updated on 2026-04-12.

Use Cases

• Train models to predict geographic coordinates based on street-level photos mentioned in the description
• Benchmark geolocation accuracy for multimodal AI systems using the provided ground-truth coordinate format
• Fine-tune models to act as OSINT/geospatial experts as instructed by the system prompts

Strengths

• Contains ground-truth coordinates in a structured XML format (<direct_lon_lat_output>) for each sample
• Data is formatted as a JSON array of ChatML messages, providing a structured conversational context

Limitations

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

Provenance

Source

huggingface

Freshness

Last updated 2026-04-12 05:41:55