Introduction to Machine Learning for Vision Applications

This workshop takes a hands-on approach to exploring some of the fundamental concepts of machine learning applied to computer vision. It walks through:

• Using Marimo
• Working with an image dataset
• Training an image classifier
• Comparing different models
• Running a model live on a camera feed

Quick start

You can access the workshop material directly in your web browser at this link:

➡️ https://epfl-center-for-imaging.github.io/vision-workshop/

Local installation

Make sure to have a working Python setup. Install the packages listed in requirements.txt:

pip install -r requirements.txt

Then, from the command-line, run the Marimo notebook in edit mode:

marimo edit intro_to_ml.py

You should be able to access the notebook locally at: http://localhost:2718/.

Scripts

Acquiring a training set of images

To extract a training dataset of labelled digits from images captured with a camera device, use:

• preprocessing.py

This script will save the training dataset in a local dataset folder. The class labels are inferred from the predetermined positions of the digits in a 3 X 4 grid.

Testing the model live on a camera

To extract digits and classify them based on the trained Scikit-learn pipeline, use:

• inference.py

This script will reload a Scikit-learn pipeline.pkl file, run the pipeline on the video feed from a camera device and display the results.

License

This workshop material is distributed under the terms of the BSD-3 license.