Hello,
I am trying to use the Openpose pretrained model with 2 cameras, for live inference on both simultaneously. Is it possible to do this without loading two copies of the model into memory, and how would I go about doing so?
Thanks for the help,
Felix
Hi, @felix.chippendale,
Codelets are assumed to run single-threaded and independently. If the resource is really under stress guess we can connect both input channels to the same codelet for inference and “de-mux” the output channel. The downside is:
1, the incoming stream of data might have out-of-order timestamp;
2, double input rate so half processing time;
3, no load balancing so in case of dropping data due to message queue overflow, there is no guarantee to have fair dropping across input channels;
4, De-muxing might be tricky;
Hope this helps.
Hello yangl,
thank you for your advice and insight.
Kind regards,
Felix