DJI-ThermalVis: An All-Time Multimodal Dataset for Urban Drone-Based Small Dim Target Detection

DJI ThermalVis is a visible light thermal infrared dual-mode object detection dataset designed to provide massive all day and multi scene remote sensing data for ground object detection and related algorithm evaluation from the perspective of unmanned aerial vehicles in urban environments. This dataset contains 13 pairs of video sequences and 5457 pairs of visible light thermal infrared image pairs, covering various scenes such as streets, highways, and squares under all-weather conditions at all times. The DJI ThermalVis dataset provides over 360K object detection annotations for six types of target objects, including cars, electric vehicles, and pedestrians, with approximately 80% of the target sizes being less than 50 pixels, making it highly challenging.