Geo-CoT380k: A Large-Scale Remote Sensing Vision-Language Dataset for Geospatial Reasoning

Geo-CoT380k is a large-scale dataset introduced in the RSThinker project to support faithful reasoning in remote sensing vision-language models. It contains structured GeoSpatial Chain-of-Thought annotations, moving beyond simple final answers. The dataset was created by author 'minglanga' and was last updated on the platform in May 2026.

Use Cases

• Training vision-language models for geospatial reasoning based on structured Chain-of-Thought annotations.
• Benchmarking model performance on faithful and verifiable reasoning tasks in remote sensing.
• Developing models that can explain their reasoning process for interpreting satellite or aerial imagery.

Strengths

• Designed for a specific research goal: supporting faithful and verifiable reasoning in vision-language models.
• Contains structured GeoSpatial Chain-of-Thought annotations, which suggests a detailed reasoning format.

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.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface

Collection Method

Introduced in the RSThinker project; specific collection method is not detailed in the provided description.

Freshness

Last updated 2026-05-26 06:43:02; freshness should be verified.