Shuai Yuan, Hanlin Qin, Renke Kou, XiangYan, Zechuan Li, Chenxu Peng, Dongliang Wu, Huixin Zhou

We present a novel infrared small target label generation (IRSTLG) framework named energy double guided single-point prompt (EDGSP). Experiments on both public (e.g., SIRST, NUDT-SIRST, IRSTD-1K) demonstrate the effectiveness of our method. Our main contributions are as follows:

1. To the best of our knowledge, we present the first study of the learning-based IRSTLG paradigm and introduce EDGSP creating a crucial link between label generation and target detection task.

2. We propose target energy initialization (TEI), double prompt embedding (DPE), and bounding box-based matching (BBM) strategies to address insufficient shape evolution, label adhesion, and false alarms.

3. For the first time, three baselines equipped with EDGSP achieve accurate annotation on three datasets. The downstream detection task illustrates that our pseudo label surpasses the full label. Even with coarse point annotated, EDGSP achieves 99.5% performance of full labeling.

If the implementation of this repo is helpful to you, just star it！⭐⭐⭐

• Note that using the “fixed” file to correct seven obvious errors in the raw data！！！
• SIRST3: SIRST, NUDT-SIRST, and IRSTD-1K
• Our project has the following structure:

  ├──./datasets/
  │    ├── SIRST3
  │    │    ├── images
  │    │    │    ├── XDU0.png
  │    │    │    ├── XDU1.png
  │    │    │    ├── ...
  │    │    ├── masks
  │    │    │    ├── XDU0.png
  │    │    │    ├── XDU1.png
  │    │    │    ├── ...
  │    │    ├── Centroid
  │    │    │    ├── XDU0.png
  │    │    │    ├── XDU1.png
  │    │    │    ├── ...
  │    │    ├── masks_coarse
  │    │    │    ├── XDU0.png
  │    │    │    ├── XDU1.png
  │    │    │    ├── ...
  │    │    ├── img_idx
  │    │    │    ├── train_SIRST3.txt
  │    │    │    ├── test_SIRST3.txt
  
  

python train_LG_SCTrans.py

python test_LG_SCTrans_PdFa.py

*This code is highly borrowed from IRSTD-Toolbox. Thanks to Xinyi Ying.

*The overall repository style is highly borrowed from DNA-Net. Thanks to Boyang Li.

If you find the code useful, please consider citing our paper using the following BibTeX entry.

@ARTICLE{10902427,
  author={Yuan, Shuai and Qin, Hanlin and Kou, Renke and Yan, Xiang and Li, Zechuan and Peng, Chenxu and Wu, Dongliang and Zhou, Huixin},
  journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing}, 
  title={Beyond Full Labels: Energy-Double-Guided Single-Point Prompt for Infrared Small Target Label Generation}, 
  year={2025},
  volume={18},
  number={},
  pages={8125-8137},
  keywords={Annotations;Shape;Object detection;Training;Adhesives;Transformers;Manuals;Labeling;Image segmentation;Feature extraction;Infrared small target label generation (IRSTLG);interactive segmentation;single-point prompt;target detection},
  doi={10.1109/JSTARS.2025.3545014}}

Welcome to raise issues or email to yuansy@stu.xidian.edu.cn for any question regarding our EDGSP.