Taxonomies and inter-class relationships are not explicitly supported in Hasty. You will be able to add attributes for objects themselves. But attributes in relation to another annotation get tricky. As there is no specification that this “specific attribute is talking about that other specific attribute”. So when you export the data, and an image has 5 balls and 5 people, how do you determine which ball is being acted on by which person? The same problem applies to the attribute predictor in Hasty. Since it doesn’t have the inter-label inter/intra-class relational context, it won’t be able to predict these attributes very reliably to help you out.
Something that’s simpler to do and might work in some limited scenarios is:
Create a new class. Use this class to create simpler, separate labels that contain all the objects that can have relations with each other. So in the example above, you’d create a box around the ball and the person. And then add attributes to this one label while which has the full context (like an attribute: things: [person, ball], action: [kicked], or however you’d like). After export, you can easily parse this class’s labels with the knowledge that they contain hierarchy information in the attributes.
We will make note of the lack of support for taxonomies in the tool. Maybe we can plan for it in our product road-map, it’s always good to get some feedback. But I hope this may be useful.