This project is developed for our MICCAI 2022 paper: ShapePU: A New PU Learning Framework Regularized by Global Consistency for Scribble Supervised Cardiac Segmentation.


  1. The MSCMR dataset with mask annotations can be downloaded from MSCMRseg.
  2. Our scribble annotations of MSCMRseg have been released in MSCMR_scribbles.
  3. The scribble-annotated MSCMR dataset used for training could be directly downloaded from MSCMR_dataset.
  4. The ACDC dataset with mask annotations can be downloaded from ACDC and the scribble annotations could be downloaded from ACDC scribbles. Please organize the dataset as the following structure:

  -- TestSet/
  -- train/
  -- val/


  1. Set the “dataset” parameter in, line 55, to the name of dataset, i.e., “MSCMR_dataset”.
  2. Set the “output_dir” in, line 57, as the path to save the checkpoints.
  3. Download the dataset, for example, the MSCMR_dataset. Then, Set the dataset path in /data/, line 110, to your data path where the dataset is located in.
  4. Check your GPU devices and modify the “GPU_ids” parameter in, line 61. Start training with:


This code has been tested with Python 3.8.5 torch 1.7.0 torchvision 0.8.0

If you have any problems, please feel free to contact us. Thanks for your attention.


View Github