🚀 PolarFree: Polarization-based Reflection-Free Imaging – [CVPR 2025]

🌟 A Solution and Dataset for Polarization-based Reflection-Free Imaging

Image source: ThinkLucid

---

🎥 Demo Video

Watch the demonstration of PolarFree in action:

Alternatively, you can play the embedded video below:

<iframe width="560" height="315" src="https://www.youtube.com/embed/g-aNjbygiJM" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>

---

📌 Highlights

✅ Large-Scale Dataset: PolaRGB includes 6,500 well-aligned RGB-polarization image pairs, 8× larger than existing datasets.
✅ Innovative Method: PolarFree leverages diffusion models to generate reflection-free priors for accurate reflection removal.
✅ State-of-the-Art Performance: Outperforms existing methods by ~2dB in PSNR on challenging real-world scenarios.
✅ Open Source: Code and dataset are freely available for research and development.

⏳ Timeline

• ✅ 2025-03-23 - 🛠️ Repository initialized with documentation.
• ✅ 2025-03-23 - 🔗 Project Page officially launched.
• ✅ 2025-03-23 - 📄 Paper available on arXiv.
• ✅ 2025-04-21 - 🚀 Provide core codebase, testing subset, and pre-trained models for evaluation.
• ✅ 2025-06-28 - 📦 Release the full PolaRGB dataset with download links.
• ✅ 2025-06-28 - 📝 Publish training code.
• TODO: Provide detailed training and testing instructions.

📖 Overview

PolarFree addresses the challenging task of reflection removal using polarization cues and a novel diffusion-based approach. Key contributions include:

• PolaRGB Dataset: A large-scale dataset with diverse indoor and outdoor scenes, providing RGB and polarization images.

• Diffusion Model: Utilizes diffusion processes to generate reflection-free priors, enabling precise reflection removal and improved image clarity.
  

• Superior Results: Extensive experiments on the PolaRGB dataset show that PolarFree outperforms existing methods by ~2dB in PSNR, achieving cleaner reflection removal and sharper image details.

• Real-World Effectiveness: PolarFree demonstrates robust performance in real-world scenarios, such as museums and galleries, effectively reducing reflections while preserving fine details.

---

🚀 Installation & Usage

1. Clone the Repository

git clone https://github.com/mdyao/PolarFree.git
cd PolarFree
pip install -r requirements.txt

2. Download the Dataset

You can access the dataset from Hugging Face:
👉 https://huggingface.co/datasets/Mingde/PolaRGB

Download and organize the dataset according to the structure required by the codebase.

Note: Currently, only the test dataset is available. The training dataset is being organized. Stay tuned!

3. Run the Demo

Once everything is set up, run the demo script:

python simple_test.py -opt options/test/test.yml -gpu_id 0

---

📊 Results

PolarFree achieves superior performance compared to existing methods:

📜 Citation

If you find this work useful, please cite:

@inproceedings{polarfree2025,
title={PolarFree: Polarization-based Reflection-Free Imaging},
author={Mingde Yao, Menglu Wang, King-Man Tam, Lingen Li, Tianfan Xue, Jinwei Gu},
booktitle={CVPR},
year={2025}
}