This project is the implement of Self-Supervised Learning-Based Framework for Speckle Reduction of Optical Coherence Tomography Images Using Frame Interpolation.

In this repo, I add an extra convex upsampling to replace previous SAFMN upsampling module to achieve a more consistent training progress. To further enhance the training process, I upate the loss and training pipeline.

git clone https://github.com/JayJiang99/FIDenoise.git
cd FIDenoise
pip install -r requirements.txt

PKU37 OCT-R1,OCT-R2 DUKE DIOME IOVS

Update the dataset process in dataset.py, self.data_root = 'The DIR of Dataset'

python inference_denoise.py 

bash train.sh

• Release the upsampling training code
• Release the evaluation code
• Release the dataset preparation code
• Update the doc for the new training pipeline

If you think this project is helpful, please feel free to leave a star or cite our paper:

@article{jiang2025FID,
  title = {Self-Supervised Learning-Based Framework for Speckle Reduction of Optical Coherence Tomography Images Using Frame Interpolation},
  author = {Jiang, Zhiyi and Hao, Yifeng and Dai, Jing and Kwok, Ka-Wai},
  journaltitle = {Advanced Intelligent Systems},
  pages = {2500001},
  issn = {2640-4567},
  year = {2025}
}

Our code is based on the implementation of ScCov, RIFE and SAFMN. Thanks to their excellent work and repository.