Fully-implemented workflow for FCN8 (based on VGG-16)

This repository handles the full workflow for applying FCN-8 for fine-scale semantic segmentation of images in a directory.

• Publication: A past version of this repository was used to train the tensrflow model associated with the following paper:

  Amgad M, Elfandy H, ..., Gutman DA, Cooper LAD. Structured crowdsourcing enables convolutional segmentation of histology images. Bioinformatics. 2019. doi: 10.1093/bioinformatics/btz083

• Dataset: The dataset associated, which was used for model training and validation can be downloaded using the instructions provided at the BCSS repository.

• Trained model: The trained tensorflow model weights can be downloaded at this link.

0: Clone repository

• git clone https://github.com/kheffah/tensorflow_FCN8Workflow

1: Initialise submodules

• cd ./tensorflow_FCN8Workflow
• git submodule init
• git submodule update

2: Install dependencies:

First update package manager:

• apt-get update

Then run the following commands ...

• python 3+ : sudo apt-get install -y python3 python3-dev python3-pip
• re 2.2.1 : pip3 install regex
• termcolor 1.1.0 : pip3 install termcolor and... : pip3 install colored --upgrade
• datetime : pip3 install DateTime
• numpy 1.12.1 : pip3 install numpy
• scipy 0.18.1 : pip3 install scipy
• PIL : pip3 install Pillow
• matplotlib 2.0.0 : pip3 install matplotlib
• sklearn 0.18.1 : pip3 install -U scikit-learn
• GPU support for tf : sudo apt-get install libcupti-dev
• tensorflow 1.1.0 : pip3 install tensorflow-gpu or... : pip3 install --upgrade https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-1.2.1-cp34-cp34m-linux_x86_64.whl

If the tensorflow installation fails, please see the followling page: https://www.tensorflow.org/install/install_linux

• tensorflow_fcn (already included as submodule) - originally at https://github.com/MarvinTeichmann/tensorflow-fcn

3: Download VGG16 weights

• cd ./tensorflow_fcn
• wget ftp://mi.eng.cam.ac.uk/pub/mttt2/models/vgg16.npy
• cd ../

4: Run test script on sample data

• Note: sample dataset was downloaded from this source and preprocessed for illustrative purposes: http://www2.warwick.ac.uk/fac/sci/dcs/research/tia/glascontest/download/warwick_qu_dataset_released_2016_07_08.zip

• RUNSETTINGPATH="$(pwd)/sampleRun/run_script/"

• python3 Run.py train $RUNSETTINGPATH (<- this tests training mode)

• python3 Run.py predict_test $RUNSETTINGPATH (<- this tests predict_test mode -test set only-)

• python3 Run.py predict_all $RUNSETTINGPATH (<- this tests predict_all mode -train/valid/test sets-)

• python3 Run.py predict_unlabeled $RUNSETTINGPATH (<- this tests predict_unlabeled mode)

Note: press Ctrl+C ay any point to stop any of the above while it is running.

5: You're good to go!