← All docs

Collections

Collections are the curation track: you organize particles by what they are, not where they came from, to build training material.

Collection

A is a hand-curated set of particles gathered across many samples and series for a labeling purpose — “Chlorella Reference Particles,” “Debris Examples,” “Ambiguous Diatoms for Review.” You curate a collection; it grows as you annotate and add particles.

Compendium

A assembles the collections one classifier needs to tell apart, each collection acting as a class. It’s scoped to a single

so the resulting model is coherent.

Dataset

A is a frozen, versioned snapshot exported from a compendium, with fixed train/validation/test splits. Once created it’s immutable — a new version is a new dataset — which is what makes a training run reproducible.

How they differ

Collection → Compendium → Dataset. A collection is a curated pile of particles (one idea, e.g. “debris”). A compendium is the full class space for a task (debris vs Chlorella vs Scenedesmus). A dataset is a frozen snapshot of that compendium, ready to train on. Curate collections, compose a compendium, export a dataset.

A collection (grouped by identity) is also distinct from a series (grouped by provenance).

Quick start

  1. Curate s of labeled particles.
  2. Assemble them into a for your classification task.
  3. Move on to Classification to export a dataset and train.