GenEISP: Generalizable Neural Electromagnetic Inverse Scattering

This repository contains the PyTorch implementation of the paper "GenEISP: Generalizable Neural Electromagnetic Inverse Scattering".

🌐 Project Page | 📄 Paper | ▶️ Video

TODO :

• Release training dataset and train code
• Release data processing code

1. Introduction

We propose the first generalizable, physics-driven inverse scattering framework, which works in an end-to-end manner and achieves more accurate reconstructions on unseen cases. It remains robust even with a single transmitter and achieves real-time inference with over 300,000x speed-up.

2. Preparation

Environment Setup

Create and activate conda environment:

conda create -n GenEISP python=3.11.10
conda activate GenEISP
pip install -r requirements.txt

Dataset Setup

1. Download the test datasets
   Please download the test datasets from GoogleDrive.

2. Extract and organize the data
   Unzip the downloaded files and place them in the data/test/ directory. The folder structure should look like this:

   GenEISP/
   ├── data/
   │   └── test/
   |       ├── 3Dmnist_test/
   │       ├── cylinder/
   │       ├── mnist/
   │       └── IF/
   

   • cylinder/, mnist/, and IF/ correspond to different test datasets.
   • Each subfolder contains the relevant test samples.

3. Training data
   The training dataset and training code will be released soon. Please stay tuned for updates.

Pre-trained Model Setup

1. Download the pre-trained model
   Download the pre-trained model weights from GoogleDrive.

2. Organize the model weights
   Unzip the downloaded files and place them in the model/ directory.

3. Evaluation

⚙️ 3.1. Configuration File (config.yaml)

Configuration files follow the naming convention:
[dataset_name]_noise[level]_N[transmitter_number].yaml, where each part indicates the experimental setup:

Example File	Components	Meaning
cylinder_noise05_N16.yaml	cylinder + noise05 + N16	Cylinder dataset, 5% noise, 16 transmitter
IF_FDE_noise00_N8.yaml	IF + FDE + noise00 + N8	Real-world IF dataset, FoamDielExt case, 0% noise, 8 transmitters
IF_FDI_noise00_N8.yaml	IF + FDI + noise00 + N8	Real-world IF dataset, FoamDielInt case, 0% noise, 8 transmitter
IF_FTD_noise00_N18.yaml	IF + FTD + noise00 + N18	Real-world IF dataset, FoamTwinDiel case, 0% noise, 18 transmitters
mnist_noise30_N16.yaml	mnist + noise30 + N16	MNIST dataset, 30% noise, 16 transmitters
mnist_noise05_N1.yaml	mnist + noise30 + N1	MNIST dataset, 30% noise, 1 transmitters

Note:
If you want to change the number of transmitters, you can modify experiment.channels in the configuration yaml file.
For example:

experiment:
  channels: [0, 1, 2, 3]  # Use 4 transmitters, represent channel 0, 1, 2, 3

The length of the channels list determines the transmitter number.

▶️ 3.2. Run Evaluation

To evaluate the model on your chosen dataset, run one of the following commands:

• For 2D datasets (e.g., cylinder, mnist, IF):

python main.py --config your_path_to_config --mode test

• For 3D datasets (e.g., 3Dmnist):

python main.py --config your_path_to_config --mode 3Dtest

4. Training

Coming Soon.

Citing

If you find this work useful for your research, please consider citing:

@article{cheng2025generalizableneuralelectromagneticinverse,
  author    = {Yizhe Cheng and Chunxun Tian and Haoru Wang and Wentao Zhu and Xiaoxuan Ma and Yizhou Wang},
  title     = {Generalizable Neural Electromagnetic Inverse Scattering},
  journal   = {arXiv preprint arXiv:2506.21349},
  year      = {2025}, 
}