FAIR1M: Fine-Grained Object Detection in High-Resolution Satellite Imagery

FAIR1M is a dataset of 15,000 high-resolution (0.3-0.8 meter) RGB satellite images for fine-grained object detection. The images were taken by Gaogen (GF) satellites and extracted from Google Earth, and are annotated with rotated bounding boxes for objects in 5 categories. The dataset was uploaded by author 'blanchon' to Hugging Face and was last updated on December 6, 2023.

Use Cases

• Train object detection models for ships, vehicles, and airplanes based on high-resolution satellite imagery.
• Develop fine-grained recognition systems for infrastructure like courts and roads from geospatial data.
• Benchmark model performance on rotated bounding box annotations for objects in complex backgrounds.

Strengths

• Contains 15,000 high-resolution satellite images.
• Provides annotations for 5 distinct object categories (ships, vehicles, airplanes, courts, roads).
• Uses rotated bounding boxes, which are suitable for objects in aerial imagery.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified as the last update was 2023-12-06.

Provenance

Source

Images taken by Gaogen (GF) satellites and extracted from Google Earth.

Collection Method

Collection of satellite imagery and manual annotation.

Freshness

Last updated 2023-12-06 12:38:09