An infrared small target dataset and we name it IST-A.
One should donwload all files successfully before unzipping.
If you fail to unzip on ubuntu (we do have the problem, too) , try again on windows.

All images and labels of IST-A have been uploaded. Anyone who needs can take it for research.
After getting some suggestions, we decide not to upload any manuscript to the open platform. If you have to cite the dataset, you may just need to cite the website (https://github.com/SeaHifly/IST-A).

Code link: https://github.com/SeaHifly/Infrared-Small-Target
email: xuhai_0513@foxmail.com

================================================================================================================================
Basic Information of IST-A.

Images in IST-A shares the same size of 288×384. Birds, UAVs, helicopters, planes, and aircraft which can fly at a relatively far distance would be taken as infrared small targets in IST-A. There are 59553 images in training subset while thare are 14756 images in testing subset.There are some images without any target because the targets may be occluded by some other objects like buildings or vegetation. In IST-A, over 87% of targets are smaller than 6×6. Although we focus on targets in the sky, it should be emphasized that the ground is also a major part of the background in IST-A. Vegetation, buildings, and some other objects on the ground are all involved in IST-A.

There are no synthetic images in IST-A. All images are captured by cameras and all targets in IST-A are noncooperative. (Targets are not controlled by us and all targets are caputured by chance.) There are not more than three targets in one image.

Attention or Notes: We only focus on targets in the sky, which means objects on the ground won't be taken as targets in IST-A.

=================================================================================================================================
Label Information

L cx cy w h

size on pixel scale:
L cx*384 cy*288 w*384 h*288

All images are labeled on the Matlab Platform. So when you are adopting IST-A on Pycharm or VS, you have to decrease the cx and cy by one on pixel scale.

On Matlab:  x∈[1,384]
On Pycharm: x-1∈[0,383]


======================================================================================================================================
If you find the dataset is helpful for your work, please cite the following paper:

H. Xu, S. Zhong, T. Zhang and X. Zou, "Multi-scale Multi-level Residual Feature Fusion for Real-time Infrared Small Target Detection," in IEEE Transactions on Geoscience and Remote Sensing, doi: 10.1109/TGRS.2023.3269092.

@ARTICLE{10106465,
  author={Xu, Hai and Zhong, Sheng and Zhang, Tianxu and Zou, Xu},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={Multi-scale Multi-level Residual Feature Fusion for Real-time Infrared Small Target Detection}, 
  year={2023},
  volume={},
  number={},
  pages={1-1},
  doi={10.1109/TGRS.2023.3269092}}