Skip to content

Tutorials

Walk-throughs that teach a concept by working through a runnable example. Each tutorial below links to a notebook in notebooks/use_cases/ and/or a packaged pipeline or trainrun YAML shipped with cuvis-ai, runnable with restore-pipeline / restore-trainrun.

Tutorials are grouped by training style:

  • Statistical — pipelines that learn from background statistics alone, no gradient steps. Fast to train, interpretable, strong baselines.
  • Gradient — pipelines that include trainable parameters fit by backpropagation. More expressive, more compute.

Running these locally

Notebook-based tutorials live in notebooks/use_cases/. See Running the Notebooks for the repeatable recipe to provision an environment (base install, per-notebook plugins, JupyterLab). Pipeline/trainrun tutorials need only the base install plus the restore-pipeline / restore-trainrun CLI, as shown on each tutorial page.

Statistical

  • RX Anomaly Detection

    Classical Mahalanobis-distance anomaly detector. Start here for hyperspectral anomaly fundamentals.

  • Channel Selection

    Two-phase training that learns which wavelengths to keep from a hyperspectral cube.

  • Blood Perfusion

    NDVI-style two-band differential rendering blood perfusion as a false-RGB overlay.

Gradient

  • Deep SVDD

    One-class anomaly detection that learns a compact representation of "normal" data.

  • AdaCLIP

    Vision-language anomaly detection coupling a frozen CLIP backbone with a trainable adapter.

  • Concepts → Training — the two-phase training model behind every cuvis-ai pipeline.
  • Workflows — task-recipe "I want to…" guides once you know what you're doing.
  • Datasets catalog — the demo datasets each tutorial runs against.