DATransNet: Dynamic Attention Transformer Network for Infrared Small Target Detection

Official implementation of paper "DATransNet: Dynamic Attention Transformer Network for Infrared Small Target Detection". Our paper is accepted in GRSL.

Network Structure

Requirements

• Python 3.8
• Windows10, Ubuntu18.04 or higher
• NVDIA GeForce RTX 4080
• Pytorch 1.13.0
• More details from requirements.txt

Dataset

Or you can download in Baidu Cloud with code of "mxhe".

Commands for Training

• Run train.py to train our network

  Python train.py
  

Cited by

《Adaptive Strategies for Multiscale Gradient Fusion in Neural Networks》 indicates that our network is suitable for the tasks of visual light targets detection.

《Hybrid attention and adaptive feature fusion network for infrared small target detection》 shows a good balance in computation cost and precision among other models

Citation

@ARTICLE{10947728,
  author={Hu, Chen and Huang, Yian and Li, Kexuan and Zhang, Luping and Long, Chang and Zhu, Yiming and Pu, Tian and Peng, Zhenming},
  journal={IEEE Geoscience and Remote Sensing Letters}, 
  title={DATransNet: Dynamic Attention Transformer Network for Infrared Small Target Detection}, 
  year={2025},
  volume={},
  number={},
  pages={1-1},
  keywords={Feature extraction;Transformers;Data mining;Training;Object detection;Image edge detection;Head;Measurement;Geoscience and remote sensing;Artificial intelligence;Infrared small target detection (ISTD);convolution neural network (CNN);Dynamic Attention Transformer;global feature extraction},
  doi={10.1109/LGRS.2025.3557021}}

Chinese Introduction

The chinese introduction is accessiable at https://blog.csdn.net/weixin_45358930/article/details/147562104?spm=1001.2014.3001.5501.

Weights

We could offer the weights for IRSTD-1K Weight_for_IRSTD_1K and NUDT-SIRST weight_for_NUDT_SIRST.