7 - Inherit a new dataset version

Objectives:

• Create easily a dataset version
• Transfer customized way annotations from one version to another

Fork a new dataset version from an existing one

It is highly recommended to leverage the dataset versioning system in order to keep track of the work done on a dataset. To create a new version from the initial dataset, select all or some assets from the initial one and click “Create new version” to initialize a new dataset version containing the selected assets. Depending on your needs, you can choose the elements from the initial version you want to embed in the new one among asset tags, labels, and annotations.

To fork a dataset with the SDK:

Transfer labels & annotations from one version to another

If you want to import the annotations but with some personalization inside the labels or annotations to be forked, you can first just fork your dataset without copying labels or annotations. Once the new dataset version is created, go to its "Settings" tab. First, you can import all or a bench of classes from any other existing version. For sure after import, you can add some extra classes also.

For sure you can also create brand new labels by selecting "New Labels"

Now that your new dataset version labels are set, you can personalize the import of the annotations from another dataset version. To do so, you can go to the “Settings” > “Annotations” part of the new dataset version. You can select the dataset version from which you want to copy the annotations, then you just
need to select class by class the annotations to inherit from the first version to the new one. It allows you to copy annotations even if the labels' name changes and merge annotations from two or more classes in one.

👍

Globally all those dataset management tools are aimed at helping data scientists create collaboratively, efficiently, and with traceability the perfect dataset for their future model training!

You can still create a new dataset version at any moment using the Python SDK: