Pytorch implementation for TMI paper DLTTA: Dynamic Learning Rate for Test-time Adaptation on Cross-domain Medical Images, by Hongzheng Yang, Chen Cheng, Meirui Jiang, Quande Liu, Jianfeng Cao, Pheng-Ann Heng, Qi Dou.

In this repository, we provide the implementation of our dynamic learning rate method on OCT dataset. The ATTA and Tent implementation were adopted from their official implementation. (Tent, ATTA)

To reproduce results on Camelyon17 and Prostate datasets, please refer to the experiments folder.

The OCT dataset can downloaded from here.

The Camelyon17 dataset can be downloaded from here.

The Prostate dataset can be downloaded from here.

1. create conda environment

   conda create -n DLTTA python=3.7
   conda activate DLTTA
   

2. Install dependencies:

   1. install pytorch==1.7.0 torchvision==0.9.0 (via conda, recommend)

3. download the dataset

4. download the pretrained model from google drive

5. modify the corresponding data path and model path in test.sh

6. run test.sh to adapt the model

If this repository is useful for your research, please cite:

@article{2022DLTTA,
 title={DLTTA: Dynamic Learning Rate for Test-time Adaptation on Cross-domain Medical Images},
  author={Hongzheng Yang, Cheng Chen, Meirui Jiang, Quande Liu, Jianfeng Cao, Pheng Ann Heng, Qi Dou},
  year={2022}
}  

Please feel free to contact '[email protected]' if you have any questions.