Hello, thanks for your nice works. After reading the ATL paper I am confused about the affordance features. You said in the paper

We first extract the human, object, and affordance features via the ROI-Pooling from the feature pyramids.

What are these affordance features actually? I mean from where these features are pooled?