Quick Start

This is a set of simple tests to use DeepReg command line tools. More details and other options can be found in Command Line Tools.

First, install DeepReg and change current directory to the root directory of DeepReg.

Train a registration network

Train a registration network using unpaired and labeled example data with a predefined configuration:

deepreg_train --gpu "" --config_path config/unpaired_labeled_ddf.yaml --log_dir test

where:

  • --gpu "" indicates using CPU. Change to --gpu "0" to use the GPU at index 0.

  • --config_path <filepath> specifies the configuration file path.

  • --log_dir test specifies the output folder. In this case, the output is saved in logs/test.

Evaluate a trained network

Once trained, evaluate the network using a test dataset:

deepreg_predict --gpu "" --ckpt_path logs/test/save/weights-epoch2.ckpt --mode test

where:

  • --ckpt_path <filepath> specifies the checkpoint file path.

  • --mode test specifies prediction on the test dataset.

Warp an image

DeepReg provides a command line interface (CLI) tool to warp an image/label with a dense displacement field (DDF):

deepreg_warp --image data/test/nifti/unit_test/moving_image.nii.gz --ddf data/test/nifti/unit_test/ddf.nii.gz --out logs/test_warp/out.nii.gz

where:

  • --image <filepath> specifies the image/label file path.

  • --ddf <filepath> specifies the ddf file path.

  • --out <filepath> specifies the output file path.