Mask R-CNN for nucleus Detection and Segmentation (in progress)

This source includes nucleui segmentation code using Mask-RCNN.

Code was implemented on ubuntu 16.04, python3.5, CUDA9.0, and tensorflow1.12.0.

Dataset

Download preprocessed images for segmentation

   mv downloaded_images  MASK_RCNN_ROOT/datasets/nucleus

(optional) Download original nucleus dataset

Model

Download pretrained model

   mv downloaded_model  MASK_RCNN_ROOT/logs/nucleus20190130T0908/

(optinal) Data preparation

To make input data in Mask-RCNN, a python file below separates overrapping mask in a original mask image using erosion and dilation algorithm (keneral size 7x7).
Also, it reduces image size by half (1024x1024) and changes format from tiff to jpg to reduce memory load

   cd MASK_RCNN_ROOT/samples/nucleus
   python3 make_patch(erosion-dliation).py

Training/Testing

   cd MASK_RCNN_ROOT/samples/nucleus
   python3 nucleus_training.py
   python3 nucleus_testing.py

Current Result

TODO

separate image into several parts, segment indepentantly.

Citation

@misc{matterport_maskrcnn_2017,
  title={Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow},
  author={Waleed Abdulla},
  year={2017},
  publisher={Github},
  journal={GitHub repository},
  howpublished={\url{https://github.com/matterport/Mask_RCNN}},
}