/Pytorch-Nuclei-Instance-Segmentation-Demo

This is a Pytorch demo of nuclei instance segmentation

Primary LanguagePythonMIT LicenseMIT

Pytorch-Nuclei-Instance-Segmentation-with-Watershed

This is a Pytorch demo for nuclei instance segmentation.

This demo reproduces and improves the Deep Interval-Masker-Aware Networks (DIMAN) and Marker-controlled Watershed method. The original implementation of DIMAN is showed by appiek.

In this work, we keep the preprocessing matlab code and marker-controlled watershed code in original implementation. We changed the other parts to python and pytorch version.

The validation dataset is TNBC nuclei segmentation dataset. You can also use the dataset after our preprocessing, which can be downloaded here.

Visualization

image

Packages version

  • Pytorch 1.3.1
  • numpy 1.17.3
  • pillow 6.2.1
  • optparse 1.5.3
  • scipy 1.3.1
  • opencv 4.2.0
  • SimpleITK
  • skimage 0.16.2

Quick start

  1. Download the preprocessed data from link
  2. Train the model
python train.py