Hi,
Yes the models used by the ZED SDK are specific and proprietary, unfortunately.
I understand this may not be realistic or convenient to label a new dataset but the only solution, for now, would be to capture the data from the ZED2 to be able to benchmark it against others.