Airborne Object Tracking Dataset with 5.9M Images and 3.3M Annotations

A collection of 4,943 flight sequences, each around 120 seconds long, recorded at 10 Hz in diverse conditions. It contains over 5.9 million images and more than 3.3 million 2D annotations of airborne objects, with all objects in labeled frames being annotated. The dataset was contributed by Amazon and is hosted on AWS Open Data.

Use Cases

• Train object detection models based on the 3.3 million+ 2D annotations.
• Develop and benchmark multi-object tracking algorithms using the sequential flight data.
• Test model robustness in diverse conditions as mentioned in the description.
• Analyze object appearance and behavior from the 120-second flight sequences.

Strengths

• Large scale with 4,943 sequences and over 5.9 million images.
• High annotation volume with over 3.3 million 2D labels.
• Explicit labeling of all airborne objects in frames that contain them.
• Includes 3,306,350 frames without labels, which can be used for negative sampling or background analysis.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• The geographic and temporal coverage of the flight sequences is unspecified.

Provenance

Source

Amazon

Collection Method

Collected as flight sequences at 10 Hz in diverse conditions.

Time Range

unknown

Freshness

unknown

Geography

unknown