Deepsea-Detector

A deep-sea creature detector for the 2022 MATE Machine Learning Satellite Challenge.

Developed by Peyton Lee, Neha Nagvekar, and Cassandra Lam as part of the Underwater Remotely Operated Vehicles Team (UWROV) at the University of Washington.

Overview

Deepsea-Detector is built on MBARI's Monterey Bay Benthic Object Detector, which can also be found in FathomNet's Model Zoo. The model is trained on data from NOAA Ocean Exploration and FathomNet, with assistance from WoRMS for organism classification. All the images and associated annotations we used can be found in our Roboflow project.

Deepsea-Detector uses YOLOv5 for detections and Norfair for tracking.

Installation

You'll need Python 3.8 or higher for this project. We use Git LFS to store some of our larger files, including the weights for our YOLOv5 model.

Open a command terminal and run the following lines:

git clone https://github.com/ShrimpCryptid/deepsea-detector.git
cd deepsea-detector
git lfs install; git lfs fetch; git lfs pull
pip install -r requirements.txt

This downloads and installs this project and all of its dependencies.

Running Inference

Running via Deepsea-Detector UI

Run the following command to boot up the UI:

python src/deepsea_detector_ui.py

This will boot up the Deepsea-Detector UI! Follow the prompts to set up your output video and output comma separated value (CSV) files.

Deepsea-Detector will output a CSV file that includes the first timestamp/frame an organism appeared, its predicted classification, and the timestamp/frame of its exit from view. It will also output an MP4 video file that shows localizations (and optionally classifications as well).

Running via Command Line Interface (CLI) or Google Colab

Our UI is a wrapper around the Deepsea-Detector CLI. This allows you to process multiple videos in sequence and define an output directory for videos. To see the full list of options and arguments, run:

python src/detection.py --help

You can also see an example in our Google Colab notebook, which gives you access to GPU resources for free!

Additional Resources:

Dataset: Roboflow project\ Model Training: Google Colab Python Notebook\ Additional Documentation: In-depth Project Explanation

Acknowledgements:

We would like to thank the following people and organizations:

• MATE and NOAA Ocean Exploration for hosting this competition,
• The World Register of Marine Species (WoRMS), FathomNet, and NOAA Ocean Exploration for providing data,
• our mentor, Rick Rupan,
• and to Professor Joseph Redmon for his guidance and support.