Bellybutton Image Segmentation (bellybuttonseg)

This is a machine-learning-based image segmentation package, designed for ease of use. Bellybutton uses a convolutional neural network, trained on images and paired user-generated masks, and outputs individually ID'd regions. Using the package requires no writing code; by editing a text file, users can edit hyper-parameters, modify data augmentation (e.g. rotate images), specify outputs and filetypes, and run training and prediction.

Quickstart guide

Install Bellybutton from command line / terminal

Create a new python environment, here we named it 'bellybuttonenv' (e.g. using anaconda: shorturl.at/elRTZ)

conda create --name bellybuttonenv

Activate this environment (e.g. using anaconda)

conda activate bellybuttonenv

Install dependencies

pip install numpy
pip install matplotlib
pip install tensorflow
pip install opencv_contrib_python==4.5.5.64
pip install scipy
pip install scikit-image 

Download Bellybutton package

pip install bellybuttonseg

Run Bellybutton on your own dataset

• Create a directory and bellybutton-required folders. When the code line below is run, you will be prompted to select a location and then prompted to enter a name for this project ([INPUT] Enter folder name for new project:). The entered name will be the name of the base folder.

python3 -c 'from bellybuttonseg import createdir; createdir()'

• Move your images, masks, and (optionally) areas of interest into the appropriate folders created in this new directory.

• Edit the parameters inside the parameters.txt file if desired.

• Run Bellybutton on your data. You will be prompted to select your project folder after running the below command.

python3 -c 'from bellybuttonseg import runBB; runBB()'

Results will be stored inside a new folder inside [yourproject]/outputs/. You may always edit the base_parameters.txt file in your base folder and run Bellybutton again with the new parameters using the above line.

Enjoy!

Run Bellybutton on an example dataset (not yet working)

Create the example directory. When this line is run, you will be prompted to select a location for the example dataset to be saved.

python3 -c 'from bellybuttonseg import createdir; createdir(example=1)'

Note that the directory you selected now has a new folder: example_project_[DATASETINFO]. Inside there folders that hold images (train, test, predict), masks, areas of interest, and outputs. We have chosen [INSERT DATASET INFO]...

[optional] Edit hyper-parameters by opening and editing [new directory]/base_parameters.txt. A complete list of parameters and their meanings is included further in this readme.

Run Bellybutton on this dataset. When this line is run, you will be prompted to select the base folder for your project, in this case select example_project_[DATASETINFO] that you just created.

python3 -c 'from bellybuttonseg import runBB; runBB()'

The algorithm should run and you can track its progress on the command line. A new folder (whose name is based on the date and time) will be created inside example_project_[DATASETINFO]/outputs/ that will store results. Once training and prediction are completed, peruse the results inside this folder. Note that a new used_parameters.txt file has been created inside this outputs folder that stores the parameters used.

To run Bellybutton again using different parameters, edit the base_parameters.txt before executing the above code line again.

Adjustable Parameters

[to be filled out]