$\Phi$-GAN: Physics-Inspired GAN for Generating SAR Images Under Limited Data

1.Introduction

This project is for paper $\Phi$-GAN: Physics-Inspired GAN for Generating SAR Images Under Limited Data.

1.1 Features

The proposed $\Phi$-GAN framework overview.

1.2 Contribution

• A physics-inspired GAN framework, $\Phi$-GAN, is proposed for SAR image generation, aiming to improve training stability and generalization under data-scarce conditions.
• $\Phi$-GAN consists of a physics-inspired neural module for PSC parameter inversion and two specialized physical loss functions for training regularization.
• Extensive experiments are conducted on diverse SAR image generation tasks. Built upon multiple existing conditional GAN architectures, the proposed $\Phi$-GAN consistently demonstrates strong adaptability, improved generalization, and robust generation performance.

2.Getting Started

2.1 Data

MSTAR dataset is used in the experiments. Dictionary and pre-trained weights for physics-inspired neural module can be downloaded from the link https://drive.google.com/drive/folders/1Yl_eupJBl1P_1CNB9q4i4xXlxJbPPRvc?usp=sharing.

2.2 Training

To train a $\Phi$-GAN model, run the following command:

python train.py \
   --bs 32 \
   --lrg 0.0001 \
   --lrd 0.0001 \
   --num_epochs 2000 \
   --save_dir ${SAVE_PATH} \
   --train_txt 'train.txt'\
   --d_mat 'D_80*80_image_domain.mat'\
   --d_h_mat 'D_80*80_image_domain_H.mat'\
   --inv_d_mat 'D_80*80_image_domain_Inv_norm.mat'\
   --f_est 'HQS_epoch_30.pth'

2.3 Generating

After training stage, run the following command to generate SAR target images with given label and angle information.

python generate.py

3.Citation

If you find this repository useful for your publications, please consider citing our paper.

@inproceedings{zhang2025ph,
  title={$\Phi$-GAN: Physics-inspired GAN for generating SAR images under limited data},
  author={Zhang, Xidan and Zhuang, Yihan and Guo, Qian and Yang, Haodong and Qian, Xuelin and Cheng, Gong and Han, Junwei and Huang, Zhongling},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={29075--29085},
  year={2025}
}