Examining Joint Demosaicing and Denoising for Single-, Quad-, and Nona-Bayer Patterns

SaiKiran Tedla, Abhijith Punnappurath, Luxi Zhao, Michael S. Brown
Samsung AI Center Toronto & York University

📄 Paper (PDF) 📄 Dataset

📌 Citation

If you use our dataset or code, please cite:

@inproceedings{Tedla2025ExaminingDemosaic,
  title={{Examining Joint Demosaicing and Denoising for Single-, Quad-, and Nona-Bayer Patterns}},
  author={{Tedla, SaiKiran and Punnappurath, Abhijith and Zhao, Luxi and Brown, Michael S}},
  booktitle={{Proceedings of the IEEE International Conference on Computational Photography (ICCP)}},
  year={{2025}}
}

---

🚀 Getting Started

This section describes how to train and test our unified demosaicing and denoising model.

🔧 Environment Setup

conda env create -f environment.yml
conda activate examine_demosaic

Please copy the DNG provided in the dataset link (into PyTorch/utilities). This DNG is primarily used for visualization and also computing the DeltaE metric.

You'll also need a Weights & Biases (wandb) account for experiment tracking. Update the wandb settings in the YAML files accordingly.

📂 Dataset

Download the dataset and then update config files accordingly with the appropriate paths.

🏋️‍♂️ Training

To train the model, run:

python PyTorch/runner.py --config configs/unified_train.yaml

Modify the config file to:

• Choose the desired ISO levels
• Set the correct paths to training patches and full-resolution images
• Update your wandb project and entity details

---

🧪 Testing

To test the model, run:

python PyTorch/runner.py --config configs/unified_test.yaml

Again, be sure to:

• Set the appropriate ISO levels
• Provide the correct paths to the test dataset
• Provide the path to the appropriate checkpoint file (see models)
• Please set plot_images:True in the config file if you want to get npy and png outputs.

---