Airborne Disaster Video Dataset with 32 Annotated Features

32 features across 5 categories like Environment and Damage annotate public videos of natural disasters. The dataset was used for the TRECVID DSDI task from 2020-2022 and is maintained by the National Institute of Standards and Technology. All footage consists of airborne, low-altitude video from disaster events.

Use Cases

• Train video classification models using the 32 features, such as Environment and Damage, for scene understanding.
• Benchmark video retrieval systems for disaster response against the TRECVID DSDI task ground truth.
• Analyze co-occurrence patterns of annotated features like Vehicles and Infrastructure in disaster footage.

Strengths

• Features are organized into 5 distinct, high-level categories: Environment, Vehicles, Water, Infrastructure, and Damage.
• Dataset is actively maintained for the TRECVID benchmark, with a last update recorded in February 2023.
• All videos provide a consistent perspective as they are airborne and captured from a low altitude.

Limitations

• Dataset size, row count, and specific video duration are not provided, limiting assessment of scale.
• The 'other-license-specified' requires careful review for permissible use cases before download.
• Content is geographically and thematically focused on natural disaster events, which may limit generalizability.

Provenance

Source

National Institute of Standards and Technology

Collection Method

Videos and annotations compiled for the TRECVID DSDI (Disaster Scene Description and Indexing) task.

Time Range

Associated with the TRECVID task spanning 2020-2022.

Freshness

Updated as of 2023-02-23; described as continually updated for the ongoing TRECVID task.

Geography

null