Skip to content

PyTorch volume toolkit. Efficient data loading, dataset conversions, visualization tools

License

Notifications You must be signed in to change notification settings

torchvtk/torchvtk

Repository files navigation

torchvtk

PyTorch volume data loading framework

Installation Instructions

The latest GitHub release is pushed to PyPi:

pip install torchvtk

To get the master run:

pip install git+https://github.com/torchvtk/torchvtk.git@master#egg=torchvtk

Optional for DICOM stuff only:

conda create --name "tvtk" python=3.6 && conda activate tvtk
conda install gdcm -c conda-forge
pip install pydicom dicom_numpy h5py numpy matplotlib

If you need DICOM, and thus gdcm, your Python version needs to be <=3.6 Modify tvtk in the third line (both after --name and at the end of the line) to your preferred environment name or just add the required packages to your existing environment. Note that the restriction to Python <= 3.6 is due to gdcm and higher version should work as well if you don't need DICOM loading capabilities.

Creating the Numpy Files for the HDF 5 File generation.

hdf5/nifiti_crawler.py This script generates out of the nifti Files of the Medical Decathlon Challenge numpy arrays that contain the images and the segmentation groundtruths.

Creating the HDF5 Files

hdf5/hdf5_crawler.py This script generates hdf5 Files with different compression techniques. These files contain the image as well as the groundtruth. The image is normalized between [0,1] and is stored in the Float32 Format. The Groundtruth is saved as an Int16 Format. The used compressions are gzip, szip and lzf.