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

KFGOD provides 882,399 annotated object instances across 33 fine-grained classes from KOMPSAT-3 and KOMPSAT-3A satellite imagery. The dataset offers annotations in both oriented and horizontal bounding box formats, serving as a homogeneous benchmark for remote sensing tasks. It was created by user ohhan777 and last updated on Hugging Face in January 2026.

Use Cases

• Train object detection models based on the 33 fine-grained classes.
• Benchmark remote sensing algorithms using the large-scale, annotated instance data.
• Compare model performance on oriented versus horizontal bounding box annotations.
• Develop applications for analyzing high-resolution satellite imagery with a ground sample distance of 0.55–0.7 meters.

Strengths

• Large scale with approximately 882,399 annotated object instances.
• High-resolution source imagery with a ground sample distance of 0.55–0.7 meters.
• Dual annotation format providing both oriented and horizontal bounding boxes.
• Homogeneous benchmark design focused on fine-grained object detection.

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 timestamp is in the future (2026-01-18).

Provenance

Source

KOMPSAT-3 and KOMPSAT-3A satellite imagery.

Collection Method

Annotations created for object detection tasks; specific methodology not detailed.

Freshness

Last updated 2026-01-18 20:16:13