An-Unsupervised-Learning-Bidirectional-3D-Image-Registration

The model contains 1 encoder and 2 decoders for bidirectional (source -> target, target -> source) 3D images registration with skip connections like Unet.

How many GPUs do you need?

• You have to set at least 4 GPUs in config.py file

Keras with Tensorflow backend

• Please intall it when you want to run.
• Don't forget to set your checkpoints and logs folders

How to run?

• python train.py
• python test.py

You can visualize your training process using Tensorboard

• tensorboard --logdir=/logs

Acknowledgement

• This model is quite similar to Balakrishnan, Guha, et al. "An Unsupervised Learning Model for Deformable Medical Image Registration." arXiv preprint arXiv:1802.02604 (2018) paper.

Note: You can only use it for research purposes.