I have been asked to look into developing a deep learning tool to identify objects from 360 degree panoramas. Can this be done using the deep learning tools in ArcGIS Pro? I suppose there are two issues here. First is it suitable for images not at a nadir angle and secondly are stitched panoramas a suitable source. Very new to all of this.
In Esri's world non-nadir imagery is referred to as oriented imagery. This video on Driving Intelligence from Video and Oriented Imagery at Scale Using AI in ArcGIS (presenters David Yu, Shairoz Sohail, and Rob Fletcher) from the Esri 2020 Developer's Summit talks about one example of how oriented imagery could be used with AI in ArcGIS. This should answer the first part of your question - can the Esri AI tools be used with non-nadir imagery. I don't know enough about the topic to be able to answer the second part of your question regarding stiched 360 degree panoramas.