Getting Started

This guide will help you set up your environment and understand the tools we’ll be using throughout the course.

Required Tools and Accounts

Core Tools

• Python: We’ll be using Python 3.8 or higher

• Anaconda/Miniconda: For managing Python environments and packages

• Git: For version control and accessing course materials

• VSCode (recommended) or PyCharm: For code editing and development

• Image Annotation Software: For creating training datasets

• Roboflow (Easiest and most intuitive, but not a great long-term solution): This web-based annotation platform was used in the pilot class associated with our course.

  • Create a free account at roboflow.com

  • Supports multiple annotation types (bounding boxes, polygons, keypoints)

  • Easy export to common formats (YOLO, COCO, VOC)

  • Includes data augmentation features

• BIIGLE: Open-source annotation tool designed with marine use in mind.

  • Create a free account at biigle.de

  • Supports multiple annotation types

  • Can be self-hosted

  • Ideal for annotating marine imagery

  • Great community

  • Alternative Open Source Options:

    • LabelImg: Simple, desktop-based tool for bounding boxes

    • CVAT: Powerful open-source annotation tool

      • Can be run locally or deployed as a server

      • Supports multiple annotation types

      • Good for team collaboration

    • VGG Image Annotator (VIA): Browser-based tool

      • No installation required

      • Works completely offline

      • Supports points, lines, polygons, and bounding boxes

    • Label Studio: Open source data labeling tool

      • Supports multiple data types beyond just images

      • Can be self-hosted

      • Has both community and enterprise versions

Required Accounts

• Google Account: For accessing Google Colab

• Hugging Face Account: For accessing and sharing models

  • Join the OceanCV organization on Hugging Face to access our models and contribute your own

  • Request to join by visiting the organization page and clicking “Request to join”

  • This will allow you to collaborate on models and datasets with other course participants

• GitHub Account: For accessing course materials and contributing

Working Environments

Option 1: Google Colab (Recommended for Beginners)

Google Colab provides a free, cloud-based Jupyter notebook environment with GPU support. To use, this is the simplist way to work in the website oceancv.org, especially when displaying resources to large classes:

1. Navigate to any .ipynb file in the course

2. Click the “Open in Colab” button at the top of the notebook

3. Save a copy to your Google Drive to preserve your work

Option 2: Local Setup

To run the course materials locally:

1. Clone the repository:

git clone https://github.com/atticus-carter/cv.git
cd cv

2. Create and activate a conda environment:

conda env create -f environment.yml
conda activate cv

3. Launch Jupyter Lab:

jupyter lab

Option 3: Local Book Build

To build and view the complete book locally:

1. Install the required dependencies:

pip install -r requirements.txt

2. Build the book:

jupyter-book build .

3. View the book by opening _build/html/index.html in your browser

Self-Teaching Guide

If you’re using this book for self-study rather than as part of a formal class, here’s how to make the most of it:

Lesson Structure

Most lessons in this book follow a consistent structure:

1. Overview: A brief introduction to the topic and its relevance to marine computer vision

2. Learning Objectives: Clear statements of what you should be able to do after completing the lesson

3. Content: Core concepts and explanations, often with visual examples

4. Code Examples: Practical implementations demonstrating the concepts

5. Exercises: Mini challenges to test your understanding

6. Additional Resources: Links to further reading and related topics

Types of Lessons

The book contains several types of content:

1. Markdown (.md) files: Conceptual lessons focused on explaining principles and ideas

2. Jupyter Notebooks (.ipynb): Interactive lessons with executable code that you can run and modify

3. Project Files: Complete workflows demonstrating real-world applications

Progressive Learning Path

The book is organized into chapters that build upon each other:

• Chapter 1: Introduces marine imaging and AI fundamentals

• Chapter 2: Covers computer vision basics like image annotation and augmentation

• Chapter 3: Dives into specific CV techniques (classification, object detection, etc.)

• Chapter 4: Guides you through a synthesis project applying what you’ve learned

Working Through Mini Exercises

Each lesson includes mini exercises that reinforce learning. For most effective self-teaching:

1. Read through the lesson: Understand the concepts before attempting exercises

2. Try exercises independently: Attempt to solve problems before looking at solutions

3. Experiment with code: Modify examples to deepen understanding

4. Complete end-of-lesson challenges: These consolidate what you’ve learned

Implementation and Practice

For most beneficial self-learning:

1. Maintain a personal project: Apply techniques to your own marine imagery dataset

2. Create implementation notebooks: Document your experiments and findings

3. Build incrementally: Start with simpler techniques before attempting advanced ones

4. Review and troubleshoot: Debug issues to build problem-solving skills

Using External Resources

Supplement your learning with:

1. Documentation: Refer to official libraries documentation (PyTorch, Tensorflow, etc.)

2. Academic Papers: Many lessons reference original research papers

3. Community Forums: Sites like Stack Overflow for troubleshooting

4. Video Tutorials: For visual explanations of complex concepts

Remember that computer vision is a hands-on field - the more you practice implementing these techniques, the better you’ll understand them.

Development Tools

Hugging Face Integration

We’ll be using Hugging Face for model sharing and deployment. Key components:

• Transformers Library: For accessing and using pre-trained models

• Datasets Library: For accessing and sharing datasets

• Model Hub: For sharing trained models with the community

Jupyter Notebooks

• Most lessons are in Jupyter notebook format (.ipynb)

• Can be run in Colab, locally, or viewed in the book

• Include both explanatory text and executable code cells

How to Get Help

1. Course Issues: Use the GitHub repository’s Issues page

2. Direct Contact: Email attcart@uw.edu for any help or questions related to the course

3. GitHub Discussions: Ask questions on the GitHub repository’s Discussions tab for community support

4. Technical Problems:

   • Check the Tools Troubleshooting section in the book

   • Ask questions in the GitHub Discussions

   • Open an issue with detailed error information

Contributing

We welcome contributions to improve the course materials! Here’s how:

1. Report Issues

   • Use the GitHub Issues page

   • Include detailed description and steps to reproduce

   • Add relevant labels (bug, enhancement, documentation, etc.)

2. Suggest Improvements

   • Open a discussion on GitHub

   • Describe your proposed changes

   • Link to relevant issues or materials

3. Submit Changes

   • Fork the repository

   • Create a new branch for your changes

   • Make your changes following our style guidelines

   • Submit a Pull Request with a clear description

   • Link relevant issues