This is the official website of the LMOD dataset.

⭐ The dataset application is very simple and requires only the following two steps:

• Please fill in this application form.
• Please send your completed application form to this E-mail address:rs_devotee@163.com.
  When we receive your application, we will reply as soon as possible. Thank you for your support!

• The LMOD dataset is the first satellite video moving multi-object detection dataset with both large-scale and multiclass labeling features. LMOD consists of eight sequences from seven videos.

• LOMD has a wide range of annotation, the smallest image width is 1500×1160, and the largest image width is 4000×2000. The large range of scenes can better simulate the effect of object detection methods used in real scenes, but at the same time, it brings more challenges for object detection.
  

• The LMOD is labeled with 459,713 vehicle objects, 9,390 aircraft objects, 10,536 ship objects and 693 train objects, for a total of 480,332 objects, with each sequence labeled with at least two classes of objects.

• Satellite videos used in LMOD are collected from JiLin-1 satellite constellation and ISS(International Space Station).

📫 If you have any questions, please contact rs_devotee@163.com.

If you want to use multi-object detection and tracking or single-object tracking dataset labeled with the OBB (Oriented Bounding Box) method which has orientation information, you can try to use the OODT dataset that we have published before.

If our work is helpful to your research, please consider citing it. We appreciate your support！

@ARTICLE{11142571,
  author={Zhang, Yuwei and Tang, Yuqi and Liu, Fengqing and Chen, Yuzeng and Han, Te},
  journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing}, 
  title={Structural Consistency Based Multiobject Detection for Satellite Video and Benchmark With Real Scene Simulation Dataset}, 
  year={2025},
  volume={18},
  number={},
  pages={23728-23743},
  doi={10.1109/JSTARS.2025.3603130}}