ROAD-UAE Features

Continuous Footage

Each video runs for 2.5 to 3 minutes, contributing to a total of 57 minutes of video in the dataset.

Heterogenous Annotations

Three main categories of agents are annotated and: people, vehicles, and traffic lights. The vehicle category is divided into 6 classes and people into pedestrians and cyclists.

Variety of Scenarios

Captures the unique road topology and traffic congestion characteristic of the Gulf Region.

Tracking Annotations

Bounding boxes with consistent tracking ids are provided for all road agents over the frames.

Action Annotations

Action labels generated as ”action tubes” at the frame level.

Road Event Annotations

Road event detections in the form of triplets (Agent, Action, Location) labeled as tubes of frame-level detections.

ROAD-UAE Stats

Total number of annotated frames: 34 386

Total number of boundind boxes annotations: 277 150

Total number of unique agents: 4209

Total duration of the videos: 57 min

Total number of vehicles: 3729

Total number of pedestrians: 222