Using label spec for model predicted labels?

[weiji14]

Hi there,

Just found out about this through the Cloud Native Geospatial talk and had a question to understand the scope of this specification. Is this spec meant for training labels only, or can it be applied to model-generated labels (e.g. the output of a machine learning process)? Specifically, I'm looking at Item Properties -> label:methods and seeing 'automated' or 'manual', so would the 'automated' option be ok to use for model predicted labels?

For example, say there is an object detection model that detects trees in an image (output bounding boxes). Would it be ok to:

1. store those predicted bounding box labels as a STAC item and use this Label extension?
2. have some sort of link to the method/code that generated the labels?