PGNet

📁 Data Preparation

1. Download the Datasets

First, please download the required datasets from the following links:

• SIRST: Download Link
• NUDT-SIRST: Download Link
• IRSTD-1k: Download Link

2. Organize the Directory Structure

After downloading and extracting the files, place them in the datasets folder located in the project's root directory. Your project should adhere to the following structure:

PGNet/
├── datasets/
│   ├── IRSTD-1K/
│   │   ├── images/
│   │   │   ├── XDU0.png
│   │   │   └── ...
│   │   ├── masks/
│   │   │   ├── XDU0.png
│   │   │   └── ...
│   │   ├── train.txt
│   │   └── test.txt
│   │
│   ├── NUDT-SIRST/
│   │   ├── images/
│   │   │   ├── 000001.png
│   │   │   └── ...
│   │   ├── masks/
│   │   │   ├── 000001.png
│   │   │   └── ...
│   │   ├── train.txt
│   │   └── test.txt
│   │
│   └── SIRST/
│       ├── images/
│       │   ├── Misc_1.png
│       │   └── ...
│       ├── masks/
│       │   ├── Misc_1.png
│       │   └── ...
│       ├── train.txt
│       └── test.txt
│
├── train.py
└── test.py

---

💡 Usage

1. Training

Run the following command to start training the model. The script will automatically load the data from the datasets directory.

python train.py

We also provide pretrained model weights, which can be downloaded from the link below:

• Pretrained Models: Baidu Pan Link (Password: dq4b)

2. Testing

Use the test.py script to evaluate the model's performance. Please ensure you have either downloaded the pretrained models or completed the training process.

python test.py

For your convenience, we have also provided the prediction maps for the three datasets:

• Prediction Maps: Baidu Pan Link (Password: iigd)

---

🙏 Acknowledgements

The codebase for this project is heavily borrowed from the IRSTD-Toolbox. We would like to express our sincere gratitude to the original author, Xinyi Ying, for their excellent work.