TopAir: Simulated Aerial Video Frames with Depth and Segmentation Annotations

TopAir provides video frames captured from a nadir view using the AirSim simulator in various outdoor environments. The dataset includes corresponding depth maps, semantic segmentation maps, and extrinsic camera locations for each frame. It was curated by Yara AlaaEldin and shared by MaLga Lab, with a last recorded update on 2025-11 18.

Use Cases

• Train monocular depth estimation models based on the provided depth map annotations.
• Develop semantic segmentation models for aerial scenes based on the segmentation map annotations.
• Research sim-to-real transfer for UAV perception tasks based on the simulated AirSim environments.
• Explore multi-task learning for joint depth and segmentation prediction based on the paired annotations.

Strengths

• Includes three aligned annotation types per frame: RGB image, depth map, and semantic segmentation map.
• Provides extrinsic camera location data for each frame, which could support 3D reconstruction tasks.
• Covers diverse simulated outdoor environments, including both rural and urban settings.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data is simulated, which may introduce a domain gap when applying models to real-world imagery.

Provenance

Source

MaLga Lab, via Hugging Face.

Collection Method

Collected using the AirSim simulator with a nadir view in various outdoor environments.

Time Range

null

Freshness

Last updated 2025-11-18 18:48:23.

Geography

Simulated environments, both rural and urban; specific real-world geography is not specified.