EMT: A Visual Multi-Task Benchmark Dataset for Autonomous Driving in the Arab Gulf Region

--This paper introduces the Emirates Multi-T ask (EMT) dataset - the first publicly available dataset for autonomous driving collected in the Arab Gulf region. It contains over 30,000 frames from a dash-camera perspective, along with 570,000 annotated bounding boxes, covering approximately 150 kilometers of driving routes. The EMT dataset supports three primary tasks: tracking, trajectory forecasting and intention prediction. Each benchmark dataset is complemented with corresponding evaluations: (1) multi-agent tracking experiments, focusing on multi-class scenarios and occlusion handling; (2) trajectory forecasting evaluation using deep sequential and interaction-aware models; and (3) intention benchmark experiments conducted for predicting agents' intentions from observed trajectories. The dataset is publicly available at avlab.io/emt-dataset, and pre-processing scripts along with evaluation models can be accessed at github.com/A S autonomous driving technology advances, the ability of data-driven models to generalize across diverse road environments and conditions is essential for safe operation, but remains a significant challenge. To achieve robust generalization, it is critical to train models on datasets that capture a wide range of traffic scenes and characteristics. Current autonomous driving datasets provide extensive coverage of regions like the USA [1-5], Europe [6, 7], and parts of Asia, including China and Singapore [1, 8]. However, the Arab Gulf region, with its unique driving conditions, remains underrepresented. To address this gap, we introduce the Emirates Multi-Task (EMT) dataset, collected in the United Arab Emirates (UAE) to capture the region's distinct traffic conditions. This region offers diverse driving challenges due to its range of road layouts, including expansive highways, urban areas, and complex city junctions. Additionally, driving behavior in the UAE reflects a blend of modern regulations and traditional practices. This work was supported by Khalifa University of Science and Technology under A ward No. RIG-2023-117. The annotated dataset supports multiple benchmarks, including tracking, trajectory prediction, and intention prediction, aimed at advancing models robustness in complex driving environments. The tracking benchmark dataset is designed to evaluate the ability of algorithms to accurately identify and maintain consistent object tracking over time in a complex driving environment. Similar to current state-of-the-art (SOT A) tracking benchmarks [1, 9, 10], it focuses on the motion of vehicles, pedestrians, cyclists, and motorbikes, captured from a frontal camera perspective. The benchmark is designed to test tracking models under varying levels of traffic congestion and frequent lane changes. The dataset contains 8,806 unique tracking IDs, including 8,076 vehicles, 568 pedestrians, 158 motorbikes and 14 cyclists, and with a mean tracking duration of 6.5 seconds.