Create a Dataset

Required Steps

• Create a new dataset under Training>Training View.

• Define a name

• Define your label query for the dataset. Use the defaults or decide what labels to train on and what labels to use as background. Per default, all project labels are selected for training, except for the background label which selected to be used as background.

• Use the defaults or specify which of your sources you want to include as train sources, which sources you want to include as test sources and which sources you do not want to use in your dataset

• If you decided to apply the additional curation process and you accepted annotations, you can specify whether you want to include only accepted annotations or keep the box unchecked if you want to use both accepted and unreviewed annotations. Excluded annotations are never included into datasets by default.

Optional Steps

• You can view the details of your dataset once it is finished building in the Training>Training view. From here it can be useful to make sure you have included all the labels you intended the model to recognize, that all the labels have enough data, get tips on which labels may need some more data collection to achieve better results, and so on. You can get this information by clicking on the Detail button for the dataset. You can also find information whether a lot of your data is smaller than the image input size of 224x224 pixels.

If at any point you need help from the Passio team, please reach out to us at [e-mail]