ssivalenka / hashi-implementation

Python implementation of HASHI paper

HASHI-Implementation

Python implementation of HASHI paper

To use this implementation, the following depencies are needed:

• Python 2.7
• Numpy
• scipy
• PyTorch 1.0
• CUDA tookit 9 (for GPU computation)
• PIL (Python Image Library)

The implementation runs as a single Python file and takes as command line arguments the following:

• Path to the directory containing the WSI’s
• Path to the directory of the the extracted tiles
• Path to the text file to which softmax probabilities of the sampled tiles are written.

To run: python full_hashi_algo_iter.py path_to_wsi_dir path_to_sampled_tiles path_to_softmax_textfile

Place the trained CNN model to be used in HASHI scheme in the models_cnn directory.

Currently, CNN models built in PyTorch are supported.