Quantifying Nvidia Xavier's performance

Hi,

I am interested in doing image classification(opencv), and deep speech work in my project. I have two wide angle 4k ip cameras that encodes video with h264/h265 that I would like to do image recognition with. I haven’t worked with opencv or nvida before so I am not sure what transformations are necessary though I would need to decode h264 and then the transformations and finally image classification.

The camera uses these “protocols” according to the datasheet.

Protocols TCP/IP, ICMP, HTTP, HTTPS, FTP, DHCP, CGI DNS, DDNS, RTP, RTSP, RTCP ONVIF, ISAPI (PROFILE S, PROFILE G), PSIA, GB
28181

I would like to run deep speech in the background as well 😅

Do you think xavier would be performant enough to do all these duties at once? Has anyone tried to do anything similar? I’d imagine I probably would run into bottlenecks memory wise.

Hi,
We have benchmarks in
Jetson AGX Xavier: Deep Learning Inference Benchmarks | NVIDIA Developer
FYR. And please check DeepStream SDK:
Announcing new software update for DeepStream 5.1