Add an end-to-end anomaly detection tutorial using PyTorch Lightning

[rashidrao-pk]

📚 Documentation

Motivation

PyTorch Lightning is widely used for supervised learning tasks, but there is currently no end-to-end tutorial demonstrating anomaly detection.

Anomaly detection is an important application in industrial inspection, healthcare, cybersecurity, and time-series monitoring. A beginner-friendly tutorial would help users understand how Lightning can simplify training and evaluation for these workflows.

Proposed tutorial

I would like to contribute a tutorial that demonstrates:

• Building an Autoencoder (or Variational Autoencoder) using LightningModule
• Organizing datasets with LightningDataModule
• Training only on normal samples
• Computing reconstruction-error anomaly scores
• Selecting an anomaly threshold using validation data
• Evaluating on normal and anomalous samples
• Logging metrics with Lightning loggers
• Visualizing reconstructed images and anomaly scores

The tutorial would follow Lightning's recommended coding style and be suitable for beginners.

Possible extensions

Depending on maintainer feedback, the tutorial could later include:

• Callbacks for reconstruction visualization
• TensorBoard logging
• Multiple datasets
• Industrial anomaly detection examples

I'd be happy to implement this if the maintainers think it would be a useful addition.

cc @lantiga

[ifrahgul]

@rashidrao-pk Hi! I am a student specializing in Deep Learning and I am highly interested in implementing this Anomaly Detection tutorial. I have experience with Autoencoders and would love to build this using LightningModule and LightningDataModule. Could you please assign this issue to me?