CoUDA

We provide the original implementation for “Collaborative Unsupervised Domain Adaptation for Medical Image Diagnosis (IEEE TIP 2020)”.

Dependencies

Python 2.7
Tensorflow-gpu 1.12.0
Opencv-python
Numpy 1.15.4

Usage

Please use this code after downloading the dataset and the model.

Dataset and Model

Dataset: The attached dataset is Office-31 corrupted by the label noise rate 0.1.
Please download the dataset [here] and put the dataset file (domain_adaptation_images) in the main directoty.

Model: Plase download the trained model [here], and put the model file (dual_log_office) in the main directory.

Training

Take the model adapted from webcam to amazon as an example. There are two steps:

1. Set the environment file: vi src_office/conf/local_nn_dual.yml
2. CUDA_VISIBLE_DEVICES=0 python src_office/train_dual.py >> ./dual_log_office/ours/webcam_2_amazon.txt src_office/conf/local_mn_dual.yml

Test

Take the model adapted from dslr to webcam as an example. There are two steps:

1. Set the environment file: vi src_office/conf/predict_dual.yml
2. CUDA_VISIBLE_DEVICES=0 python src_office/predict_dual.py >> ./dual_log_office/ours/dslr_2_webcam/test.txt src_office/conf/predict_dual.yml

Citation:

If you use this code and dataset, please cite:

@article{zhang2020collaborative,
  title={Collaborative Unsupervised Domain Adaptation for Medical Image Diagnosis},
  author={Zhang, Yifan and Wei, Ying and Wu, Qingyao and Zhao, Peilin and Niu, Shuaicheng and Huang, Junzhou and Tan, Mingkui},
  journal={IEEE Transactions on Image Processing},
  year={2020}
}