Brikwerk / siamese-segmentation

Image co-segmentation with a Siamese Neural Network based off of CoSegNet

Siamese Neural Network for Object Co-Segmentation

This project aims to generate predictive background/foreground masks from image pairs with similar objects. CoSegNet's architecture was used as the basis for this project.

Getting Started

First, create the "icoseg_data" directory at this project's root and then download/extract the iCoseg dataset (what I used to train, feel free to use another alternative dataset) into the folder.

Next, install all the python requirements:

pip install -r requirements.txt

To enable the tqdm progress bar in the Jupyter Notebook, please run the following command after dependencies are installed:

jupyter nbextension enable --py widgetsnbextension --sys-prefix

If you wish to train the model and evaluate it, run:

jupyter notebook

and open up the "train.ipynb" file.

The model, datasets, and other utilities can be found in the src folder.