This project was originally developed for the paper: Towards Source-Free Cross Tissues Histopathological Cell Segmentation via Target-Specific Finetuning. If you use this project in your research, please cite the following works:
@article{li2023towards,
title={Towards Source-Free Cross Tissues Histopathological Cell Segmentation via Target-Specific Finetuning},
author={Li, Zhongyu and Li, Chaoqun and Luo, Xiangde and Zhou, Yitian and Zhu, Jihua and Xu, Cunbao and Yang, Meng and Wu, Yenan and Chen, Yifeng},
journal={IEEE Transactions on Medical Imaging},
year={2023},
publisher={IEEE}
The TNBC dataset with mask annotations can be downloaded from: TNBC.
The TCIA dataset with mask annotations can be downloaded from: TCIA.
The KIRC dataset with mask annotations can be downloaded from: KIRC.
torch>=1.9.0
opencv-python>=4.5.1.10
SimpleCRF==0.1.0
matplotlib>=3.3.1
Python >= 3.6
TensorBoardX
Some basic python packages such as Numpy, Scikit-image, Scipy ......
Train the model
python train_with_UNet.py --epochs=200 --batch-size=4 --mode=Source --batch-size=2 --save-every=100
Get evaluate results images
Firstly, move the evaluate result masks to the eval folder in data folder.
Then run python canny.py
- The GatedCRFLoss is adapted from GatedCRFLoss for medical image segmentation.