L2SKNet

[IEEE TGRS] Implementation of our paper "Saliency at the Helm: Steering Infrared Small Target Detection with Learnable Kernels". Paper

Highlighting our domain-aware LLSKM, unfolding the 'Center substracts Neighbors' pattern.

Requirements

• Python 3.8
• Windows10, Ubuntu18.04 or higher
• NVDIA GeForce RTX 3090
• pytorch 1.8.0 or higher
• More details from requirements.txt

Datasets

We used the NUDT-SIRST for training and test. This dataset can be found and downloaded in: NUDT-SIRST.

Please first download this dataset and place the dataset to the folder ./data/.

• The dataset in our project has the following structure:

├──./data/
│    ├── NUDT-SIRST
│    │    ├── images
│    │    │    ├── 000000.png
│    │    │    ├── 000001.png
│    │    │    ├── ...
│    │    ├── img_idx
│    │    │    ├── test.txt
│    │    │    ├── train.txt
│    │    ├── masks
│    │    │    ├── 000000_mask.png
│    │    │    ├── 000001_mask.png
│    │    │    ├── ...
│    ├── ...

Commands for Training

• Install the environment according to requirements.txt .

• Enter the repo, and run train_device0.py to perform network training:

$ python train_device0.py --model_names L2SKNet_FPN --dataset_names NUDT-SIRST

• The model_name in our code corresponds to the model name in our paper as follows:

  L2SKNet_FPN for L2SKNet-FPN;

  L2SKNet_UNet for L2SKNet-UNet;

  L2SKNet_1D_FPN for L2SKNet-FPN*;

  L2SKNet_1D_UNet for L2SKNet-UNet*.

  Note: The 'Recip' version is on the way out.

• Checkpoints and Logs will be saved to ./log/, and ./log/ has the following structure:

├──./log/
│    ├── [dataset_name]
│    │   ├── [model_name]
│    │   │    ├── 1.pth.tar
│    │   │    ├── 2.pth.tar
│    │   │    ├── ...
│    ├── [dataset_name]_[model_name]_[time].txt

Commands for Evaluate your own results

• Run test.py to generate file of the format .mat and .png (--test_epo 200 means test with the 200th epoch model):

$ python test.py --model_names L2SKNet_FPN --dataset_names NUDT-SIRST --test_epo 200

• The file generated will be saved to ./result/ that has the following structure:

├──./result/
│    ├── [dataset_name]
│    │   ├── img
│    │   │    ├── [model_name]
│    │   │    │    ├── 000000.png
│    │   │    │    ├── 000001.png
│    │   │    │    ├── ...
│    │   ├── mat
│    │   │    ├── [model_name]
│    │   │    │    ├── 000000.mat
│    │   │    │    ├── 000001.mat
│    │   │    │    ├── ...

• Run cal_metrics.py for direct evaluation:

$ python cal_metrics.py --model_names L2SKNet_FPN --dataset_names NUDT-SIRST

• The file generated will be saved to ./result/ that has the following structure:

├──./result/
│    ├── [dataset_name]_[model_name]_[time].txt
│    ├── [dataset_name]_[model_name].mat

Commands for parameters/FLOPs and runtimes calculation

• Run t_models.py for parameters and FLOPs calculation:

$ python t_models.py

• Run t_time.py for runtimes calculation:

$ python t_time.py

Acknowledgement

We extend our sincere gratitude to Xinyi Ying and colleagues for their outstanding toolbox, BasicIRSTD (Ver. July 24, 2023). Additionally, we would like to thank Luping Zhang for his invaluable contributions to this repository.

Contact

For any questions regarding this paper or the code, please feel free to reach out to wufengyi98@163.com.

Citation

@ARTICLE{Wu_2024_TGRS,
    author    = {Wu, Fengyi and Liu, Anran and Zhang, Tianfang and Zhang, Luping and Luo, Junhai and Peng, Zhenming},
    title     = {Saliency at the Helm: Steering Infrared Small Target Detection with Learnable Kernels},
    booktitle = {IEEE Transactions on Geoscience and Remote Sensing},
    year      = {2024},
    doi       = {10.1109/TGRS.2024.3521947}
}