VOXL2 RTSP decoding fails
-
@Alex-Kushleyev Thanks for all the information Alex. Actually I got it solved, the problem was with my VOXL Power cable coming from VOXLPM. After changing this wire the reboot stopped. I am wondering if there was some loose connection which was causing the problem when GPU acceleration started over VOXL. I was monitoring the current consumption of my system with voxl-inspect-battery and I noted, that before running ML model it was 0.9 Amps and after ML acceleration it was rising to 1.2 Amps.
Few observations from voxl-inspect-cpu while ML acceleration started was temperature rising from 70 (No ML acceleration) to 80 (with ML acceleration). I am using YoloV5 over GPU and 50% GPU consumption I was able to see post acceleration. Maybe my video decoding is also hardware based with some GPU usage over there causing more GPU usage. -
@Aaky , I am glad you figured it out. Most of the time the unexpected reboots are due to some sort of power issue, that is why I originally asked about the source of power - sometimes if a Power supply is used to power VOXL2, if the power supply cannot provide sufficient current, the system will reset. But sometimes, cabling is the issue.
Alex
-
@Alex-Kushleyev One quick information I wanted to have on this feature. I have this script running over my VOXL2 and it is in its default form always decoding and publishing the frames. I want it to be smart and subscriber based whereas if any mpa subscriber once gets connected to the pipe been published only then I should start to decode. The reason been if I am not listening to this stream, it keeps consuming CPU cycles. Any quick python based sample code would work. I am searching for the same but will save my time if you can point me to right location.
-
@Eric-Katzfey Can you provide me some guidance on my above query?
-
@Aaky, I see what you are asking.
What is needed is the python binding to the C function
int pipe_server_get_num_clients(int ch), which is defined herepympa_create_pubalready is set up to return the MPA channel number, so we would just need to use that channel number of checking the number of clients.I can implement this, but it will need to wait a few days. That is a good feature to have. If you would like to try implementing it yourself, you can add the bindings to :
- https://gitlab.com/voxl-public/voxl-sdk/utilities/voxl-mpa-tools/-/blob/pympa-experimental/tools/pympa.cpp (add call to
pipe_server_get_num_clients) in a similar way how other functions are called. - add the actual python wrapper to: https://gitlab.com/voxl-public/voxl-sdk/utilities/voxl-mpa-tools/-/blob/pympa-experimental/tools/python/pympa.py
- then finally, when you call
pympa_create_pub, save the pipe id and then pass it to check for clients.
Alex
- https://gitlab.com/voxl-public/voxl-sdk/utilities/voxl-mpa-tools/-/blob/pympa-experimental/tools/pympa.cpp (add call to
