Official PyTorch implementation of our method. The full paper is available at: Paper.

A novel bootstrap representation learning method by leveraging the predictable possibility of neighboring slices. At the core of our method are a simple and straightforward dense self-supervision on the predictions of local representations and a strategy of predicting locals based on global context, which enables stable and reliable supervision for both global and local representation mining among volumes.

# Preprocessing: 
python preprocess.py
# Training:
python train.py --config=${CONFIG_NAME} --batch_size=${BATCH_SIZE} --seed=${RANDOM_SEED}
# Resume training:
python train.py --config=${CONFIG_NAME} --resume=${WEIGHTS_PATH} --batch_size=${BATCH_SIZE} --seed=${RANDOM_SEED}
# Testing: 
python eval.py --config=${CONFIG_NAME} --seed=${RANDOM_SEED} --trained_model=best --no_sort --is_test --display
# Evaluation: 
python evaluation.py --config=${CONFIG_NAME} --metric=${METRICS} --seeds=${RANDOM_SEEDS} --root=${EXP_FOLDER}

You can create a definition in <configs/*_cfg.py>, then you can use any of the training commands in the previous section. (See the comments in <config.py> for an explanation of each field):

    my_custom_config = Conifg({
        'name': '',
        'dataset': my_custom_dataset_config,
        'transformer': my_custom_augmentation_config,
        'model': my_custom_model_config,
        'loss': my_custom_loss_config,
        'optimizer': my_custom_optimizer_config,
        # anything else.
    })

if you find this code useful for your research, please cite:

@article{
  title     = {Bootstrap Representation Learning for Segmentation on Medical Volumes and Sequences},
  author    = {Zejian Chen and Wei Zhuo and Tianfu Wang and Wufeng Xue and Dong Ni},
  journal   = {CoRR},
  volume    = {abs/2106.12153},
  year      = {2021},
}

For questions about our paper or code, please contact Zejian Chen.