Jiaxiang Ren1, Kicheon Park2, Yingtian Pan2, Haibin Ling1
1Department of Computer Science, 2Department of Biomedical Engineering
Stony Brook University
This repository is the official Keras implementation of Content-Aware BMA Removal model (CABR). Paper
tensorflow == 1.14.0
keras == 2.3.1
Ensure the trained model weights is in model_weights/bestmodel.hdf5
. Then run
python -u inference.py
DICE score will be printed after inference.
Find the predicted mask and enhanced image in
.
└── dataset
├── ...
├── AwakeOCA_mask_pred_DiceXXXX.tif # Predicted mask
└── AwakeOCA_enhanced.tif # Enhanced image
The rest of dataset used in this work is not available for now due to data policy.
@inproceedings{ren2022self,
title={Self-Supervised Bulk Motion Artifact Removal in Optical Coherence Tomography Angiography},
author={Ren, Jiaxiang and Park, Kicheon and Pan, Yingtian and Ling, Haibin},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={20617--20625},
year={2022}
}
This work was supported in part by NSF Grants 1814745 and 2006665, and NIH grants R01DA029718 and RF1DA048808.