Skip to content
English
More support
Go to transkribus
  • There are no suggestions because the search field is empty.

1. Why train a Layout Recognition model?

In Transkribus, a layout model is responsible for recognising text regions, tables or baselines within a document. Training a custom layout model can significantly improve accuracy for complex or structured documents.

Unlike text recognition models, where large and diverse datasets can improve accuracy across many handwriting styles, layout models are harder to generalise across different document types. This is why you can custom train, Field Models, Table Models and Baseline Models in Transkribus.

1. Public Layout Models

Public layout models are trained on a variety of documents, making them a great starting point for general layout recognition.

  • Ready to Use – Public models are pre-trained and can be applied immediately.
  • Wide Coverage – They work well for many standard document layouts.
  • No Setup Required – Easy to use, making them ideal for quick processing.

However, layout structures vary significantly between documents, making it difficult for public models to handle all cases equally well.


2. Custom Layout Models

If public models do not provide accurate results, you can train a custom layout model to optimise layout recognition for your specific documents.

  • Higher accuracy - A custom model learns from your specific document structure, improving text region and baseline recognition.
  • Adapts to your documents - ensures multi-column text, marginal notes and annotations are processed in the correct order.
  • Improves overall transcription quality - accurate layout detection helps text recognition models produce better results.

When Should You Train a Custom Layout Model?

Custom models are particularly useful in the following situations:

  •  If your documents contain complex structures, such as newspapers, tables forms, registries or manuscripts with marginalia.
  • If public models fail to detect text regions, baselines, or reading order correctly.
  • If you work with the same type of documents regularly and need consistent results.

Public layout models are useful for general layout detection, but they may struggle with complex, historical, or highly structured documents. If you frequently correct layout issues manually, training a custom model can significantly improve efficiency and accuracy.

Click here to learn more about training Field Models, Table Models, or Baseline Models in Transkribus.