This guide provides instructions on how to install the SUIM dataset and run the associated Jupyter notebook.
The SUIM dataset is a Segmentation of Underwater IMagery (SUIM) dataset that contains ~1500 images with pixel annotations for eight object categories.
The images have been collected during oceanic explorations and human-robot collaborative experiments, and annotated by 7 human participants.
Before you begin, ensure you have the following installed on your machine:
- Python
- Jupyter Notebook
Clone the repository to your local machine:
git clone https://github.com/loomkoom/suim-segmentation.gitChange into the project directory:
cd suim-segmentationInstall the required dependencies:
python -m venv venv
.\venv\Scripts\activate
pip install -r requirements.txtDownload the SUIM dataset at: https://drive.google.com/drive/folders/10KMK0rNB43V2g30NcA1RYipL535DuZ-h
Extract the zip folders, train_val is training/validation data and test is the test data
change the ddir variable at the top of the notebook to point at the folder where you unzipped the training and test folders.
project_start has all the data exploration, preparation and early training of models up to the Convolution Neural Network. (Notebook cells with !! in front of them are needed, other cells are mostly optional).
project_models contains all the up-to-date models with the construction of the tensorflow dataset and training of the fully connected CNN's added.
project_masks contains the code to show prediction masks on full images and has some comparisons.