DIRNet-tensorflow

Tensorflow implementation of DIRNet

• [CAUTION] This implementation is actually different from the original paper. This implementation uses a simple bicubic interpolation not cubic spline interpolation. They can be different in details. If you want to more precise experiments, you need to change the codes.

Usage

# Training
python train.py

Intermediate results and model checkpoint can be found in tmp and ckpt each.

# Evaluation
python deploy.py 

Evaluation results can be found in result.

References

• End-to-End Unsupervised Deformable Image Registration with a Convolutional Neural Network
• Spatial Transformer Network
• Tensorflow implementation of STN
• Tensorflow implementation of bicubic interpolation

Author

Inwan Yoo / iwyoo@unist.ac.kr