The paper and supplementary materials for HViT can be found at:
HViT is a Hierarchical Vision Transformer model designed for medical image registration tasks. It utilizes a hierarchical vision transformer architecture to achieve accurate and efficient registration of medical images.
Please refer to the paper for detailed information on the model architecture, methodology, and experimental results.
-
Clone the repository:
git clone https://github.com/your-username/hvit.git cd hvit -
Create and activate a conda environment:
conda create -n hvit python=3.10 -y conda activate hvit -
Install PyTorch:
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121 -
Install other dependencies:
pip install -r requirements.txt
To train the model:
python src/scripts/main.py --mode train \
--batch_size 1 \
--train_data_path ./path/to/your/train \
--val_data_path ./path/to/your/val \
--num_gpus 1 \
--experiment_name "HViT_dummy" \
--max_epochs 1000 \
--lr 1e-4 \
--precision bf16
Note: You can use --num_gpus -1 to utilize all available GPUs
To resume training from a checkpoint:
python src/scripts/main.py --mode train \
--batch_size 1 \
--train_data_path ./path/to/your/train \
--val_data_path ./path/to/your/val \
--num_gpus 1 \
--experiment_name "HViT_dummy_resumed" \
--max_epochs 1000 \
--lr 1e-4 \
--precision bf16 \
--resume_from_checkpoint /path/to/your/checkpoint.ckpt
To run inference using a trained model:
python src/scripts/main.py --mode inference \
--checkpoint_path ./checkpoints/2024-09-12_17-42-10/model_epoch_10.ckpt \
--test_data_path ./path/to/your/test \
--batch_size 1 \
--num_gpus 1 \
--precision bf16
Note: Adjust the checkpoint_path to point to your trained model
If you find StablePose useful in your research, please cite our paper:
@InProceedings{Ghahremani_2024_CVPR,
author = {Ghahremani, Morteza and Khateri, Mohammad and Jian, Bailiang and Wiestler, Benedikt and Adeli, Ehsan and Wachinger, Christian},
title = {H-ViT: A Hierarchical Vision Transformer for Deformable Image Registration},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2024},
pages = {11513-11523}
}