Clarification on using the models for properties task

Hello,

I have been going through the code in the repository and I am having a hard time figuring out how to use the models for property prediction. It is also unclear to me what the task setting is in your reported results. I do not see any output head corresponding to properties in the code. What exactly is the input and output for the model in that task setting? Does the model take in an image and a polygon? An image and multiple polygons? Does it always predict some properties? I did not find any clarification in the paper or in the code.

Thank you very much!