This repository is the official Pytorch implementation of "Deep Variational Inverse Scattering" published in the European Conference on Antennas and Propagation (EUCAP 2023).
| Project Page |
(This code is tested with pytorch 1.12.1, Python 3.8.3, CUDA 11.0 and cuDNN 7.)
- numpy
- scipy
- matplotlib
- torch==1.12.1
Run the following code to install all pip packages:
pip install -r requirements.txt We used a synthetic datasets composed of 128x128 images with random ellipses. We added 30dB noise to the measurements and consider two setups for configurations of receivers and incident waves 1) full: where the sensors are uniformy distributed around the object 2) limited-view: where the sensors are located on the right side of the object. You can download the full and limited-view datasets and unzip them on the dataset folder.
This is an example of how training the model for 150 epochs for unet and 150 epochs for conditional flow models with limited-view configuration:
python3 train.py --epochs_unet 150 --epochs_flow 150 --batch_size 64 --dataset scattering --lr 0.0001 --gpu_num 0 --remove_all 0 --desc default --input_type limited-view --train_unet 1 --train_flow 1 --restore_flow 1Each argument is explained in detail in utils.py script.
If you find the code or our dataset useful in your research, please consider citing the paper.
@inproceedings{khorashadizadeh2023deep,
title={Deep variational inverse scattering},
author={Khorashadizadeh, AmirEhsan and Aghababaei, Ali and Vla{\v{s}}i{\'c}, Tin and Nguyen, Hieu and Dokmani{\'c}, Ivan},
booktitle={2023 17th European Conference on Antennas and Propagation (EuCAP)},
pages={1--5},
year={2023},
organization={IEEE}
}