Rethinking Transformer-Based Blind-Spot Network for Self-Supervised Image Denoising Paper

Usage

Datasets

Download SIDD and DND datasets, and modify dataset_path in dataset/base_function.py accordingly.

|- dataset_path
  |- SIDD
    |- SIDD_Medium_Srgb
      |- Data
        |- 0001_001_S6_00100_00060_3200_L
        |- 0002_001_S6_00100_00020_3200_N
        |- ...
    |- SIDD_Validation
      |- ValidationNoisyBlocksSrgb.mat
      |- ValidationGtBlocksSrgb.mat
    |- SIDD_Benchmark
      |- BenchmarkNoisyBlocksSrgb.mat
  |- DND
    |- info.mat
    |- images_srgb

Validation

Validate on SIDD Validation dataset,

cd validate
python base.py --config_file "../option/tbsn_sidd.json"

Training

Training on SIDD Medium dataset,

sh train.sh

Citation

If you make use of our work, please cite our paper.

@inproceedings{li2025rethinking,
  title={Rethinking Transformer-Based Blind-Spot Network for Self-Supervised Image Denoising},
  author={Li, Junyi and Zhang, Zhilu and Zuo, Wangmeng},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2025}
}