Re-labeled bounding boxes of test subset in IRDST [1].
The original/official dataset can be referred to http://xzbai.buaa.edu.cn/datasets.html.

We only annotated the bounding boxes for testing purposes, without providing the complete segmentation and bounding box labels as offered by the original authors. Our labels were annotated under local zoom conditions. Since we apply gray-level stretching during the local image display, diffused edge pixels are marked as part of the infrared small target within the regression box.

=========================
Label Information
L cx cy w h
L: class id (Unused, usually set 0)
cx, cy: target center point
w, h: target size

Convert the labels into actual coordinates in the image by:
cx*img_w cy*img_h w*384 h*img_h
img_w, img_h: the size of the image.

All images are labeled on the Matlab Platform. So when you are adopting our labels on Pycharm or VS, you have to decrease the cx and cy by one on a pixel scale.
For example (assume img_w=384):
  On Matlab:  x∈[1,384]
  On Pycharm: x-1∈[0,383]
========================

Based on our understanding, the re-labeled bounding boxes are superior to the original official box labels, which can provide a valuable reference for future researchers. It is worth noting that we previously contacted the original authors to point out certain issues with the labels, and they may have optimized some of them.

We consider the official segmentation labels [1] to be generally reasonable. When comparing the bounding boxes of these official segmentation labels with our re-labeled ones, the official labels tend to include fewer diffused edge pixels of the target. However, the assessment of which set of labels is more reasonable may vary among different researchers. Therefore, it is important to note that the bounding box labels we re-labeled are intended solely as a reference for research purposes.

We have completed a manuscript (soon to be or already submitted to TGRS), which includes some visual comparisons. If you are eager to cite these re-labeled annotations, feel free to contact us, and we may consider uploading a version to arXiv in advance. Alternatively, you can also freely cite the webpage following the relevant guidelines of any journal or conference directly.

If anyone has any questions, feel free to leave a comment or contact us via email. More information and our work will be provided in the near future.

e-mail:[email protected]

Reference:
[1] H. Sun, J. Bai, F. Yang, and X. Bai, “Receptive-field and direction induced attention network for infrared dim small target detection with a large-scale dataset irdst,” IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1–13, 2023.