No GPU or CPU cost with YOLOv5 on Jetson AGX

Hi, I been working on a project that detects objects in rtsp source using YOLOv5s.

Everything is fine, my model works perfect in the Jetson Xavier AGX. But it is too perfect because when there is no pixel change, (I mean there is no movement or any difference in the image) I observed that the GPU and CPU usage is zero. Only if there is a movement or change in the fixed camera (like security camera) image, I can observe GPU and CPU usage on jtop.

Why is this too much perfect ?? Please enlight me. Thanks!

Hi,

Which source do you use?
Is it possible that the inference is only triggered when any motion in the camera?

Thanks.

Hi, I used https://github.com/ultralytics/yolov5 repo. I don’t know is there any trigger…

Hi,

Since this is a customized source, would you mind to check with yolov5 team to see if they do inference for every frame first?

Thanks.