DFCPS

Overall structure of the DFCPS:

Model renderings:

1. Installation

• Create a conda environment:

$ conda env create -f semiseg.yaml
$ conda activate semiseg

• Install apex 0.1 (requires CUDA)

$ cd . /furnace/apex
$ python setup.py install --cpp_ext --cuda_ext

The implementation of our model code references TorchSemiSeg, which you can also refer to for further details by checking out this model.

2. Data Preparation

• Download the data. Support the import of medical image datasets, including CT scans, MRI images, etc., this time using the Kvasir-SEG dataset for segmentation, if necessary, you can replace the dataset.

• Download the pre-trained ResNet models from OneDrive link.

DATA/
|-- Kvasir
|-- pytorch-weight
|   |-- resnet50_v1c.pth
|   |-- resnet101_v1c.pth

3. Training and Inference

$ cd . /exp.voc/voc8.res50v3+.CPS+DFCPS
$ bash script.sh

• The tensorboard file is saved in log/tb/ directory.
• In script.sh, you need to specify some variables, such as the path to your data dir, the path to your snapshot dir that stores checkpoints, etc.

4. Different Partitions

To try other data partitions beside 1/8, you just need to change two variables in config.py:

C.labeled_ratio = 8
C.nepochs = 34

Please note that, for fair comparison, we control the total iterations during training in each experiment similar (almost the same), including the supervised baseline and semi-supervised methods. Therefore, the nepochs for different partitions are different.

5. Citation

@inproceedings{chen2024semi,
  title={Semi-supervised medical image segmentation method based on cross-pseudo labeling leveraging strong and weak data augmentation strategies},
  author={Chen, Yifei and Zhang, Chenyan and Ke, Yifan and Huang, Yiyu and Dai, Xuezhou and Qin, Feiwei and Zhang, Yongquan and Zhang, Xiaodong and Wang, Changmiao},
  booktitle={2024 IEEE International Symposium on Biomedical Imaging (ISBI)},
  pages={1--5},
  year={2024},
  organization={IEEE}
}