Zero-shot object recognition embedded in model?

Hello! Thank you SO much for sharing this awesome model and all it's training data.

In the Aria AR demo, the highlighted objects had a name tag to them. How was this achieved?

I would also like to know. Basically how were the values in Table 5 collected?

I am in the same situation as the others and would like to know if it's possible to get labels on the masks of what it is, ie, if it puts a mask on a dog, there's a way in the mask info to find out that it's a dog