UAV Instance Segmentation Dataset with 29,888 Aerial Tiles

UAV_INST_DATA is a dataset of 29,888 UAV (Unmanned Aerial Vehicle) image tiles with instance segmentation annotations in COCO format. It contains 114,876 annotations across 3 categories, derived from 3,802 source images. The dataset was uploaded by jeff-calderon to Hugging Face and was last updated in February 2026.

Use Cases

• Training instance segmentation models based on UAV imagery and COCO-format annotations.
• Benchmarking model performance on aerial object detection and segmentation tasks.
• Developing applications for automated analysis of objects in drone-captured scenes.
• Researching domain adaptation for computer vision models from ground-level to aerial perspectives.

Strengths

• Contains 29,888 annotated image tiles, providing a substantial volume for model training.
• Includes 114,876 instance segmentation annotations, offering detailed object-level labels.
• Provides a clear train/validation split with 26,927 training and 2,961 validation images.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset's geographic and temporal coverage is unspecified, which may limit understanding of its representativeness.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Uploaded by jeff-calderon on Hugging Face.

Collection Method

Likely collected and annotated from UAV (drone) imagery flights.

Freshness

Last updated 2026-02-06 18:35:53; freshness should be verified.