NeOTF: Guidestar-free neural representation for broadband dynamic imaging through scattering

Yunong Sun^1,*  Fei Xia^1,2,*

¹Department of Electrical Engineering and Computer Science, University of California, Irvine, CA 92697, USA
²Beckman Laser Institute, University of California, Irvine, CA 92697, USA

^*Corresponding authors

---

🚩Accepted by Advanced Photonics 2026

🔥 News

• [2026.04] 🎉🎉🎉 Congratulations! NeOTF has been accepted by Advanced Photonics.
• [2025.12] The code repo is released on Github.
• [2025.11] The preprint is available on arXiv.

🎬 Overview

NeOTF is a guidestar-free OTF retrieval method for imaging through dynamic scattering media. By optimizing a neural representation with only a few speckle images from unknown objects, NeOTF robustly retrieves the system's OTF without a guidestar.

🔧 Dependencies and Installation

1. Clone repo

   git clone https://github.com/Xia-Research-Lab/NeOTF.git
   cd NeOTF

2. Install dependent packages

   conda create -n NeOTF python=3.10 -y
   conda activate NeOTF
   pip install torch numpy pillow matplotlib tqdm pyyaml

⚡ Quick Inference

For training and reconstructing images from default multi-frame speckles, simply run:

python NeOTF.py --config ./config.yml

Run all baseline methods (HIO+ER, MORE) alongside NeOTF:

bash run_main.sh --config config.yml --output_dir ./outputs

## 📷 Results 

🧩 Repository Structure

• NeOTF.py: Main NeOTF training and reconstruction pipeline.
• MORE.py: MORE algorithm baseline.
• HIOER.py: HIO+ER algorithm baseline.
• SIREN.py: Neural network module.
• utils.py: Data loading and helper functions.
• config.yml: Default configuration file.
• run_main.sh: Benchmark bash script.

🎓 Citations

If our code helps your research or work, please consider citing our paper.

@article{sun2025neotf,
  title={NeOTF: Guidestar-free neural representation for broadband dynamic imaging through scattering},
  author={Sun, Yunong and Xia, Fei},
  journal={arXiv preprint arXiv:2507.22328},
  year={2025}
}