Introduction:

SatMTB contains 249 videos with more than 50,000 sequential frames. The object categories are divided into two levels of coarse and fine granularity, with aircraft, ships, vehicles, trains and their corresponding 14 types of fine-grained objects annotated frame by frame in the form of horizontal bounding box, rotating bounding box, masks, etc. The total number of annotation instances is 1,033,511.

SatMTB contains the significant challenges in satellite video, such as multi-task, fine-grained objects, small targets, complex backgrounds.

Sample Videos:

Download:

https://zenodo.org/records/15253996

Publications:

Li, Shengyang, Zhuang Zhou, Manqi Zhao, Jian Yang, Weilong Guo, Yixuan Lv, Longxuan Kou, Han Wang and Yanfeng Gu. “A Multitask Benchmark Dataset for Satellite Video: Object Detection, Tracking, and Segmentation.” IEEE Transactions on Geoscience and Remote Sensing 61 (2023): 1-21. doi: 10.1109/TGRS.2023.3278075.