All images are frames of the video stream from UAV and labeled by human. The bounding box is adopted in this dataset to mark the target. Each image is attached a XML file to record its information.
Other than object detection, this dataset can also be used for object classification and object tracking. The link to download the dataset: https://pitt.box.com/s/756141768nn92cj0dkfbg6dan17c4h4q