Beyond Image Prior: Embedding Noise Prior into Conditional Denoising Transformer (Condformer)

This repository is for Condformer introduced in the following paper

Huang, Y., Huang, H. Beyond Image Prior: Embedding Noise Prior into Latent Space of Conditional Denoising Transformer. Int J Comput Vis (2025). (Accepted by IJCV) arXiv Published

Dependenices

• python 3.10
• pytorch == 2.0.0
• NVIDIA GPU + CUDA

Models

Download the pre-trained models from Google Drive

Data preparing

Download DIV2K and Flickr2K datasets into the path "Datasets/Train/DF2K" for synthetic image denoising task. Download SIDD-Medium datasets into the path "Datasets/Train/SIDD_Medium_Srgb" for real image denoising task.

Settings (option.py)

For synthetic image denoising:

• '-data_train' == ['DF2K', 'WED', 'BSD']
• '-data_test' == ['CBSD68', 'Kodak24', 'Urban100']
• '-n_train' == [3450, 4744, 400]

For real image denoising:

• '-data_train' == ['SIDD_Medium_Srgb']
• '-data_test' == ['SIDD']
• '-n_train' == [320]

Train

python main.py --train 'train'

Test

python main.py --train 'test'

Calculate complexity

python main.py --train 'complexity'

Citation

@misc{huang2024imagepriorembeddingnoise,
      title={Beyond Image Prior: Embedding Noise Prior into Conditional Denoising Transformer}, 
      author={Yuanfei Huang and Hua Huang},
      year={2024},
      eprint={2407.09094},
      archivePrefix={arXiv},
      primaryClass={eess.IV},
      url={https://arxiv.org/abs/2407.09094}, 
}