Entry Point

Train

Module to train a network using init files and a CLI.

deepreg.train.build_config(config_path: Union[str, List[str]], log_dir: str, exp_name: str, ckpt_path: str, max_epochs: int = - 1) → Tuple[Dict, str, str]

Function to initialise log directories, assert that checkpointed model is the right type and to parse the configuration for training.

Parameters

  • config_path – list of str, path to config file

  • log_dir – path of the log directory

  • exp_name – name of the experiment

  • ckpt_path – path where model is stored.

  • max_epochs – if max_epochs > 0, use it to overwrite the configuration

Returns

  • config: a dictionary saving configuration

  • exp_name: the path of directory to save logs

deepreg.train.main(args=None)

Entry point for train script.

Parameters

args – arguments

deepreg.train.train(gpu: str, config_path: Union[str, List[str]], ckpt_path: str, num_workers: int = 1, gpu_allow_growth: bool = True, exp_name: str = '', log_dir: str = 'logs', max_epochs: int = - 1)

Function to train a model.

Parameters

  • gpu – which local gpu to use to train.

  • config_path – path to configuration set up.

  • ckpt_path – where to store training checkpoints.

  • num_workers – number of cpu cores to be used, <=0 means not limited.

  • gpu_allow_growth – whether to allocate whole GPU memory for training.

  • log_dir – path of the log directory.

  • exp_name – experiment name.

  • max_epochs – if max_epochs > 0, will use it to overwrite the configuration.

Predict

Module to perform predictions on data using command line interface.

deepreg.predict.build_config(config_path: Union[str, List[str]], log_dir: str, exp_name: str, ckpt_path: str) → Tuple[Dict, str, str]

Function to create new directory to log directory to store results.

Parameters

  • config_path – path of configuration files.

  • log_dir – path of the log directory.

  • exp_name – experiment name.

  • ckpt_path – path where model is stored.

Returns

  • config, configuration dictionary.

  • exp_name, path of the directory for saving outputs.

deepreg.predict.build_pair_output_path(indices: list, save_dir: str) → Tuple[str, str]

Create directory for saving the paired data

Parameters

  • indices – indices of the pair, the last one is for label

  • save_dir – directory of output

Returns

  • save_dir, str, directory for saving the moving/fixed image

  • label_dir, str, directory for saving the rest outputs

deepreg.predict.main(args=None)

Entry point for predict script.

Parameters

args –

deepreg.predict.predict(gpu: str, ckpt_path: str, split: str, batch_size: int, exp_name: str, config_path: Union[str, List[str]], num_workers: int = 1, gpu_allow_growth: bool = True, save_nifti: bool = True, save_png: bool = True, log_dir: str = 'logs')

Function to predict some metrics from the saved model and logging results.

Parameters

  • gpu – which env gpu to use.

  • ckpt_path – where model is stored, should be like log_folder/save/ckpt-x.

  • split – train / valid / test, to define the split to be evaluated.

  • batch_size – int, batch size to perform predictions.

  • exp_name – name of the experiment.

  • config_path – to overwrite the default config.

  • num_workers – number of cpu cores to be used, <=0 means not limited.

  • gpu_allow_growth – whether to allocate whole GPU memory for training.

  • save_nifti – if true, outputs will be saved in nifti format.

  • save_png – if true, outputs will be saved in png format.

  • log_dir – path of the log directory.

deepreg.predict.predict_on_dataset(dataset: tensorflow.python.data.ops.dataset_ops.DatasetV2, fixed_grid_ref: tensorflow.python.framework.ops.Tensor, model: tensorflow.python.keras.engine.training.Model, save_dir: str, save_nifti: bool, save_png: bool)

Function to predict results from a dataset from some model

Parameters

  • dataset – where data is stored

  • fixed_grid_ref – shape=(1, f_dim1, f_dim2, f_dim3, 3)

  • model – model to be used for prediction

  • save_dir – path to store dir

  • save_nifti – if true, outputs will be saved in nifti format

  • save_png – if true, outputs will be saved in png format

Warp

Module to warp a image with given ddf. A CLI tool is provided.

deepreg.warp.main(args=None)

Entry point for warp script.

Parameters

args –

deepreg.warp.shape_sanity_check(image: numpy.ndarray, ddf: numpy.ndarray)

Verify image and ddf shapes are consistent and correct.

Parameters

  • image – a numpy array of shape (m_dim1, m_dim2, m_dim3) or (m_dim1, m_dim2, m_dim3, ch)

  • ddf – a numpy array of shape (f_dim1, f_dim2, f_dim3, 3)

deepreg.warp.warp(image_path: str, ddf_path: str, out_path: str)
Parameters

  • image_path – file path of the image file

  • ddf_path – file path of the ddf file

  • out_path – file path of the output