Skip to content

AnnoTrain

Annotate. Train. Learn. Share.

AnnoTrain is an open platform for participatory scientific image annotation, deep learning, and education.

It is designed both as a research tool and as a pedagogical platform, making the complete deep-learning workflow — from raw scientific images to trained, evaluated, and shared models — accessible and understandable.

Human annotation → Dataset → Training → Evaluation → Model → Sharing → Reuse → Improvement

Status

🚧 AnnoTrain is currently under active early-stage development. This documentation describes the project's vision and architecture and will grow alongside the codebase — see the Roadmap for what exists today.

Why AnnoTrain?

Scientific image analysis is usually fragmented across disconnected tools: one application to annotate, another script to prepare a dataset, another framework to train a model, notebooks to evaluate it, and yet another platform to share the result. For students and non-specialists, this fragmentation also makes deep learning look like a black box.

AnnoTrain connects these steps into a single, understandable workflow — read more in Why AnnoTrain.

Initial goals

  • Simple and accessible participatory image annotation
  • Scientific dataset creation
  • Annotation quality control and consensus
  • Deep-learning model training and evaluation
  • Human-in-the-loop model improvement
  • Interactive visualization of predictions and model errors
  • Educational exploration of the complete deep-learning workflow
  • Open dataset and model sharing

Philosophy

AnnoTrain aims to make deep learning less of a "black box" by allowing users to participate in the entire process — from creating annotations to evaluating and improving trained models.

Researchers can use AnnoTrain to create and annotate scientific datasets, while students and other contributors can learn how human annotations become training data and how these data influence model predictions and performance.

Where to go next