Attention Based Multi-Instance Thyroid Cytopathological Diagnosis with Multi-Scale Feature Fusion

Implementation detail of our paper "Attention Based Multi-Instance Thyroid Cytopathological Diagnosis with Multi-Scale Feature Fusion".

Citation

Please cite this paper in your publications if it helps your research:

@inproceedings{qiu2021attention,
  title={Attention Based Multi-Instance Thyroid Cytopathological Diagnosis with Multi-Scale Feature Fusion},
  author={Qiu, Shuhao and Guo, Yao and Zhu, Chuang and Zhou, Wenli and Chen, Huang},
  booktitle={2020 25th International Conference on Pattern Recognition (ICPR)},
  pages={3536--3541},
  year={2021},
  organization={IEEE}
}

Requirements

• Python == 3.5
• Pytorch == 1.3.0
• torchvision == 0.4.1

Run

run main.py to start training and testing, the options are as follows:

• --dataset: dataset to train with [thyroid | breast]
• --backbone: select backbone [resnet18 | resnet34]
• --method: select method [B | BF | BFA]
• --mode: select mode [train | test]
• --save_weight: path to save checkpoint
• --load_weight: path to load checkpoint
• --random_seed: set random seed
• --gpu: gpu index

To train with our method:

python main.py --dataset thyroid --backbone resent18 --method BFA --mode train

To test with our best model:

python main.py --dataset thyroid --backbone resent18 --method BFA --mode train --load_weight ./weight/best.pth

Contact

• email: qiushuhao@bupt.edu.cn; czhu@bupt.edu.cn