V7's Neural Networks feature opens up the ability to use a trained model to automatically label your data. In this guide, we'll take a look at how you can structure your workflow to take the manual work out of labelling.
The crucial ingredient to our workflow here will be our model. If you haven't trained a model in V7 yet, follow the steps here to get one running.
If you've already created a model, head to Neural Networks, click Start, and select the minimum and maximum number of servers to use when running your model. This will determine the speed with which your model is able to label items.
Now that your model is up and running, head to the Datasets page and open the Settings page of the dataset that you'd like to label.
Add a Model stage, select which model the stage will use, and map the classes the model was trained on with the classes in the dataset.
To copy the classes from the dataset that you used to train your model directly to a new dataset, open the training dataset, click Classes, select all and click Copy to dataset.
Moving the model stage to the first stage of your workflow will reveal the option to Auto Start. This will send any new items to the model stage as soon as the workflow is saved or, in the case of an empty dataset, as soon as they are done processing.
If your workflow starts with a Model stage with Auto Start enabled, this part mostly involves kicking back, and letting the model handle the labelling in its stage or stages.
If annotation or review stages are in place before the first Model stage, annotators can trigger the model by simply sending items to the next stage.
Items can also manually be sent to the Model stage in bulk by filtering for all items in the stage prior, selecting all and clicking Advance 1 Stage.
Once the Model stage has labelled your data, your reviewers can step in to make any required changes to the model's inference.
Updated 4 months ago