Hi, it seems that the local network workflow needs some clarification.
The camera streaming is not required and it’s actually not recommended. Even so the solution from above would work, it’d be the Fusion receiver that does all the computation, which is not the spirit. The best practice is to make the sender run the AI, and then only send theire detections instead of a raw RGB image.
What you should do is run a an that just retrieves bodies and uses startPublishing on the sender side. The receiver then just need to run the fusion, and not all the cameras all over again.
See this thread for reference.