The code is for paper: MedicalMatch: Rethinking Weak-to-strong Consistency Framework from a Correlation Perspective for Semi-supervised Medical Image Segmentation.
Experiments are conducted on three public datasets: ACDC , Synapse and ISIC.
-
ACDC
We evaluate our experiments on ACDC dataset under 1% labeled, 3% labeled and 10% labeled, respectively.
More details about the dataset split and implementation details will be released till acceptance.
Refer to this link and download ACDC dataset. -
ISIC
We divide the dataset into 1838 and 756 images for training and validation, respectively. Then, we validate MedicalMatch under 3% and 10% labeled. We will upload the processed dataset later.
-
Synapse
Download from the link provided by : TransUnet.
- python 3.7
- pytorch 1.9.0
- torchvision 0.10.0
Train a Semi-Supervised Model
For example, we can train a model on ACDC dataset by:
python train_MedicalMatch.py
Then evaluate by:
python test_MedicalMatch.py
Note that all of our settings are the same with SSL4MIS .