The Label Attribute Assistant is our offering to automate adding
metadata to annotations. We do so by training an AI model based on the
data you provide. This custom-tailored approach means that you get AI
assistance that gets better over time, as you provide more and more
data.
To get started with the assistant, first, produce 25 annotations and add attributes to them.
This will trigger our model service to train the first model. This
might take 2-10 minutes, depending on the load and the quantity of data
you have per image.
When it becomes available, you will receive a notification in the bottom left corner of your screen.
Look out for this notification
You can use the assistant by first selecting an annotation. When
doing so, we will populate the attribute input(s) with suggestions - as
we do here:
Here, we get accurate attribute predictions for this nice image from the 1600's
You can then accept them by clicking on them one-by-one. Or, if you have more than one suggestion, you can accept all of them by clicking "Accept All"
Please note that as of today the attribute predictions work for the attributes of the single and multiselect types.
You can toggle the predictions on and off by pressing the toggle at the top of the Label Attributes section:
You can also check out our
MP wiki for attribution if you are interested in creating and experimenting with models.