Skip to content

Latest commit

 

History

History
38 lines (27 loc) · 1.39 KB

README.md

File metadata and controls

38 lines (27 loc) · 1.39 KB
Raw

camelyon17

Code adapted from [Camelyon17] (https://github.com/Camelyon17/camelyon17)

Requirements and Setup

  • torch
  • torchvision
  • openslide
  • opencv
  • matplotlib

Usage

Download the Camelyon17 dataset:

  • First download the label data lesion_annotations.zip, unzip it, and you will find the label data lesion_annotations/patient###_node_#.xml (where each # represents a digit) for the image data files centre_#/patient###_node_#.tif you will download.
  • From folders centre_0 ... centre_4, only download the patient###.zip files that have label files in the above folder lesion_annotations, since otherwise you will to download a much larger number of hugh image data files that cannot be used in training due to lacking of labels.

Unzip downloaded data as .tif files:

  • python3 unzip_sh.py > unzip_all.sh
  • chmod +x unzip_all.sh
  • mkdir tif
  • ./unzip_all.sh

Cut each huge .tif image file and put the results into a folder (named after the image file) containing small .png patches, and create downsized .png masks according the labels (tumor/normal) of the patches:

  • python3 make_patch.py

(Optional) If you want to see the thumbnails of the huge tif files:

  • python3 get_thumbnail.py

Make random train/dev/test splits:

  • python3 make_manifest.py

Train:

  • python3 train.py --train_epoch=10

Web server:

  • python3 main.py