PolarAnything: Diffusion-based Polarimetric Image Synthesis

Kailong Zhang^† · Youwei Lyu^† · Heng Guo^* · Si Li · Zhanyu Ma · Boxin Shi

ICCV 2025

Requirements

We test our codes under the following environment: Ubuntu 22.04, Python 3.9.23, CUDA 12.1.

1. Clone this repository.

git clone https://github.com/PRIS-CV/PolarAnything.git
cd PolarAnything

2. Install packages

conda env create -f environment.yaml

Dataset

The data is still being organized. Coming soon!

Pre-trained models

We provide the pre-trained models for inference. Just download and put them into the model folder.

Inference

• You can run inference on the example raw images using the following command:

./run_infer.sh

the results will be saved in the results/ directory

• If you want to use your own data, simply call the Python script directly with your desired parameters. For example:

python infer.py \
  --input_folder <your_input_folder> \
  --results_folder <your_results_folder> \

Train

./run_train.sh

Citation

If you find this work helpful to your research, please cite:

@misc{zhang2025polaranythingdiffusionbasedpolarimetricimage,
      title={PolarAnything: Diffusion-based Polarimetric Image Synthesis}, 
      author={Kailong Zhang and Youwei Lyu and Heng Guo and Si Li and Zhanyu Ma and Boxin Shi},
      year={2025},
      eprint={2507.17268},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2507.17268}, 
}

Acknowledgements

This code is built on Stable Diffusion and Controlnet. We thank the authors for sharing their codes.