Semantic-oriented Labeled-to-unlabeled Distribution Translation for Image Segmentation

by Xiaoqing Guo.

Summary:

Intoduction:

This repository is for our IEEE TMI paper "Semantic-oriented Labeled-to-unlabeled Distribution Translation for Image Segmentation"(知乎)

Framework:

Usage:

Requirement:

Pytorch 1.3 Python 3.6

Preprocessing:

Clone the repository:

git clone https://github.com/Guo-Xiaoqing/L2uDT.git
cd L2uDT 
bash sh_Ours.sh

Data preparation:

Dataset (Google Drive) should be put into the folder './data'. For example, if the name of dataset is CVC, then the path of dataset should be './data/CVC/', and the folder structure is as following.

ThresholdNet
|-data
|--CVC
|---images
|---labels
|---labeled.txt
|---unlabeled.txt
|---test.txt

The content of 'labeled.txt', 'unlabeled.txt' and 'test.txt' should be just like:

26.png
27.png
28.png
...

Note that we regard 'valid.txt' of EndoScene dataset as our 'labeled.txt', 'train.txt' of EndoScene dataset as our 'unlabeled.txt', and 'test.txt' of EndoScene dataset as our 'test.txt'.

Pretrained model:

You should download the pretrained model from Google Drive, and then put it in the './model' folder for initialization.

Well trained model:

You could download the trained model from Google Drive, which achieves 81.464% in Dice score on the EndoScene testing dataset. Put the model in directory './models'.

Log file

Log file can be found here

Citation:

@article{guo2020learn,
  title={Semantic-oriented Labeled-to-unlabeled Distribution Translation for Image Segmentation},
  author={Guo, Xiaoqing and Liu, Jie, Yuan, Yixuan},
  journal={IEEE Transactions on Medical Imaging},
  year={2021},
  publisher={IEEE}
}

Questions:

Please contact "xiaoqingguo1128@gmail.com"