Once you have annotated your data and decided that you are ready to train a model, the final step is to publish your dataset. This step turns your collection into a trainable, trackable, immutable dataset. You can find this dataset in your private dataset list, and from there you can train it. More details about model training can be found here.
We perform a couple validations on your data before we allow you to publish you dataset. The first is the number of annotated images. If you do not have at least 30 annotated images in your collection, we do not allow you to publish a dataset, because the resulting model will not perform well. Similarly, if you do not have at least one example of each class that is present in your dataset in each of your Train, Validation, and Test buckets, you will not be able to publish your dataset. Again, this is to ensure a successful model training session.