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

4. Managing Models

Use the Model Manager to browse public and private models, edit the metadata of your private models or publish them.

Previous step: Training Models

To access all the models, navigate to the Models workspace. Clicking on "Models", it provides a comprehensive overview of the models you have trained. You can look for models by using the filter and search options. The models are categorised into five tabs: All, Favourite, Public and My Models and Shared with me. By default, all models you train are private.

When you select one model, the model metadata appears on the right. In particular, they are: 

  • the model ID;
  • the model name; 
  • when it was trained; 
  • the creator(s) of the model; 
  • the language(s) of the document used for the training; 
  • the number of words on which it was trained; 
  • the centuries in which the material was produced;
  • the Character Error Rate on the Validation Data; 
  • the type of material (handwritten and/or printed).

By clicking the star icon to the right of the model name, you can add it to your "Favourite models" list. Your favourite models will then be suggested for quick access when you start an automatic text recognition.
The "Show Description" button provides more information about the model and the option to test it using Quick text recognition. Only if you are the user who trained the model, the "Edit" button to edit the metadata is enabled.

Make your model public

Congratulations on training your own Trankribus model Making your trained models public in Transkribus is a great way to share your work with the community. Here's how you can do it:

  1. Select the model to edit: Find the model you want to make public and click the "Edit" button (the pencil icon).

  2. Set visibility settings: On the model's edit page, scroll down until you see the "Visibility" section and choose the appropriate visibility option. You'll have three options to choose from:

    • Private: Your model will remain visible and usable only by you.
    • Public with private dataset: Your model will be visible and usable by other Transkribus users. However, the dataset used to train the model will remain private and cannot be accessed or used by others.
    • Public with public dataset: Both your model and the associated training dataset will be visible by all Transkribus users. This option allows others to not only use your model but also understand how it was trained.
  3. Click on "Submit Model for Publication". Once submitted the model will be reviewed by our team and you will be notified as soon as it has been approved. Your model will now be visible and accessible to other Transkribus users!

Important Requirements:

  • Description: Your model must have a detailed description that is more than 200 characters long. A good description helps other users understand what types of documents or scripts your model is best suited for. 

  • CER: The Character Error Rate must be 12% or lower on your evaluation set. The Character Error Rate is a crucial metric that indicates the model's accuracy. A lower CER signifies fewer errors in the recognised text, making the model more valuable for other users.

Keep in mind: 

  • Name: Give your model a descriptive name that clearly indicates its functionality or the training data it used.

    • Do: German Typewriter 1900-1930
    • Don't: German Version 3
  • Training Set Size: Aim for around 50,000 words of training data. A larger and more diverse training dataset generally leads to better model performance and accuracy across a wider range of documents.

Consider that for scripts or languages with limited existing models in Transkribus, sharing your model can still be valuable to the community, even if it has less Training Data or a CER close to 12%. Please clearly indicate the specific script/language and any limitations in the model description.

Thank you for considering sharing your models with the Transkribus community! Your contribution helps to enrich the platform and benefits other users.

 

Sharing a Model

In Transkribus, you can share a model with a certain collection. This feature makes it easier to collaborate with team members working on specific projects or collections. 

  1. Open the Model Workspace
    Navigate to your private model you want to share and click on "Show Description".

  2. Access the Sharing Options
    Click on the model to highlight it. Then, click on the "Options" menu.

  3. Select "Share with Collections"
    From the list of available options, choose Share with Collections.

  4. Choose the Target Collection
    A slide-in box will appear, allowing you to select the collection(s) you want to share the model with. Select the appropriate collection and confirm your choice.

Once shared, the model will be accessible for use within the selected collection.