Version your dataset for training or export
Dataset versions allow you to keep track of different snapshots of your dataset for export or model training. This creates a "locked" version of your dataset which you can use to train and tweak variations of your models.
You will need at least 3 annotated images in your dataset to lock a dataset version.
When a new version is created, the following attributes will be locked in your current version, and changes after that time will be applied to the next version:
- Source data
- Label definitions
- Label instructions
- Label annotations
- Label status
- 1.In your dataset, navigate to the "Versions" tab.
- 2.Click "New Version"
- 3.In the modal, click "Create Version" to lock in your current dataset.
If this is your first version, this will lock in Version 1. You will then be able to train a model with Version 1 of your dataset. Additional changes to your dataset will be applied to Version 2, and so on.