Hi, I'm using the Raspberry Pi Zero 2 W as host for the OAK-1. The RPi is running headless, connection is established via SSH and frames are forwarded via X11 to a Windows PC. [This script](https://github.com/maxsitt/insect-detect/blob/main/yolo_tracker_preview.py) is used to track detected objects. When streaming the frames via X11, **9-11 fps** can be measured, possibly due to the low computational ressources of the RPi Zero 2 and the burden of showing and streaming the frames. With the DepthAI Pipeline Graph tool **34 fps** are measured, when not showing the frames: [URL=https://i.imgur.com/JL8BSwz.png][IMG]https://i.imgur.com/JL8BSwzh.png[/IMG][/URL] If the frames are not shown and fps is printed to console, **47 fps** can be measured: ``` #cv2.imshow("tracker_preview", frame) print(f"fps: {fps}") ``` To my understanding, the lower fps when showing the frames or when using the pipeline graph tool (which output is also forwarded via X11) are caused by the higher burden on the RPi Zero 2 W, which slows down the pipeline. Is it correct that the fps measured when only printing to console are the "true" fps of the model + tracker running on the OAK-1? Thanks, Max

Reliable fps measurement on RPi Zero 2 W

maxsitt

Hi,
I'm using the Raspberry Pi Zero 2 W as host for the OAK-1. The RPi is running headless, connection is established via SSH and frames are forwarded via X11 to a Windows PC.

This script is used to track detected objects. When streaming the frames via X11, 9-11 fps can be measured, possibly due to the low computational ressources of the RPi Zero 2 and the burden of showing and streaming the frames.

With the DepthAI Pipeline Graph tool 34 fps are measured, when not showing the frames:

If the frames are not shown and fps is printed to console, 47 fps can be measured:

#cv2.imshow("tracker_preview", frame)
print(f"fps: {fps}")

To my understanding, the lower fps when showing the frames or when using the pipeline graph tool (which output is also forwarded via X11) are caused by the higher burden on the RPi Zero 2 W, which slows down the pipeline. Is it correct that the fps measured when only printing to console are the "true" fps of the model + tracker running on the OAK-1?

Thanks,
Max

jakaskerl

Hi maxsitt
It looks like that's the case, yes. The fps prints when the queue has a new frame available. The drop with pipeline graph probably stems from the constantly updating FPS. You could try running your code on a PC as well just to check.

Regards,
Jaka