M3OT Dataset Introduced for Enhanced Multi-Object Tracking with UAVs
A new dataset named M3OT has been introduced to improve multi-object tracking using Unmanned Aerial Vehicles (UAVs). This dataset is designed to address the challenges of tracking small objects in aerial imagery, utilizing both RGB and infrared thermal modalities. The M3OT dataset was acquired by two UAVs operating at high altitudes, ranging from 100 to 120 meters. It includes 21,580 frames extracted from 8 hours of video, featuring 10,790 paired RGB-IR images and 220,000 bounding boxes across various environments such as suburban, urban, daytime, dusk, and nighttime settings. The dataset aims to enhance the robustness of tracking models by providing multi-modality paired images and multi-view perspectives, which are crucial for tracking in complex scenarios. 
