After attending this session, participants will be able to:

• Understand the role of experiment tracking in Computer Vision projects and why it is important for reproducibility and model development.
• Identify the main elements that should be tracked during a Computer Vision experiment, including hyperparameters, metrics, datasets, model versions, images and predictions.
• Use MLflow to log, organize and compare Machine Learning experiments.
• Track metrics, models and artifacts generated during Computer Vision training.
• Use Weights & Biases for advanced experiment visualization and analysis.
• Log and analyze images, predictions, confusion matrices and learning curves.
• Compare experiments and identify the impact of different hyperparameter configurations.
• Understand the practical differences between MLflow and Weights & Biases and when each tool is more appropriate.
• Apply experiment tracking best practices in real Computer Vision projects and collaborative development environments.

Computer Vision projects often involve a large number of experiments, datasets, hyperparameters and model versions. Keeping track of these elements becomes increasingly important as projects grow and multiple model configurations need to be tested and compared.

“Experiment Tracking for Computer Vision” is a practical training session focused on two widely used experiment tracking tools: MLflow and Weights & Biases.

Participants will learn how to organize and monitor Computer Vision experiments, log hyperparameters and metrics, store models and other artifacts, compare different training runs and analyze model performance.

The MLflow section will focus on setting up experiments, logging results and comparing different runs through the MLflow interface. Participants will also take part in a short practical exercise by modifying a hyperparameter and observing its impact on model performance.

The Weights & Biases section will focus on more advanced visualization capabilities, including image logging, predictions versus ground truth, confusion matrices and learning curves.

The course is highly practical and includes live demonstrations and exercises designed to show how experiment tracking tools can be integrated into real Computer Vision workflows.