Using deepstream SDK for human traffic

May I have estimation? Not necessary to be very accurate.
I am doing some evaluation for a project for human traffic detection.
Say I use peoplenet from TLT and DeepStream, how many cameras one Tesla T4 GPU should be able to process for 10-15 FPS? FP16 and pruned model is quite good for me.
I just need estimate only.

you can refer to https://github.com/NVIDIA-AI-IOT/deepstream_tlt_apps#measure-the-inference-perf to measure the inference perf of your model and get the rough throughput of whole pipeline.

Can you please share any number? Could be tested with other GPUs. I just want to have a sense.

Hi @edit_or,
Sorry! We would encourage user to do the hands-on by themselves.

Thanks!