@NickM-27 commented on GitHub (Feb 4, 2024):
if you are running HA OS then you are not using tensorrt, which means you do not have the same problem as is being discussed here
@nordeep commented on GitHub (Feb 4, 2024):
Came here with the same issue: KeyError: -17
On the Frigate 0.12 I used tensor model yolov7-tiny-416 on 0.13.1 Frigate can't start with this…
@shad0wca7 commented on GitHub (Feb 5, 2024):
hhmm. In you docker config, the variable -e 'NVIDIA_VISIBLE_DEVICES'='all' I believe this needs to have your cards UUID?
"all" works within…
@hargrove4570 commented on GitHub (Feb 4, 2024):
@nordeep yes my detection is completely whack even though before it was perfect. When I have time I'm going to find the old docker release that…
@nordeep commented on GitHub (Feb 4, 2024):
@hargrove4570 for myself i decided to go back to 0.12.1 version which work perfectly or me.
@fenrisulfau commented on GitHub (Feb 5, 2024):
hhmm you docker compose looks different to mine, plus i cant see the --runtime=nvidia command anywhere? otherwise might be best deleting your…
@nordeep commented on GitHub (Feb 4, 2024):
As I can see the changes haven't been significant from 0.12.1 at tensorrt docker container - https://github.com/blakeblackshear/frigate/tree/dev/docker/…
@sobrarbe commented on GitHub (Feb 4, 2024):
Hello, I have the same problem when updating Frigate to 13, the plugin does not start. I get the following in the long. Some Help please
@hargrove4570 commented on GitHub (Feb 3, 2024):
I've been tracking these same errors with the opt/frigate/frigate/object_detection.py", line 225, I wonder if it is very sensitive to the latest…
@NickM-27 commented on GitHub (Feb 2, 2024):
will close this issue and you can create another for your use case
@sobrarbe commented on GitHub (Feb 4, 2024):
They say that going back to version 12 of the plugin works. But I don't know how to go back since it only gives me the option of the latest version in…
@hargrove4570 commented on GitHub (Feb 4, 2024):
Hey can you check you Nvidia container toolkit version with nvidia-ctk --version
Mine is 1.13.5 and I get a similar error wi th compute 5.0…
@NickM-27 commented on GitHub (Feb 2, 2024):
yes, to be clear the detect fps does not need to match the camera fps
@fenrisulfau commented on GitHub (Feb 4, 2024):
hhmm. In you docker config, the variable -e 'NVIDIA_VISIBLE_DEVICES'='all' I believe this needs to have your cards UUID?
@tkffaul commented on GitHub (Feb 2, 2024):
I was wrong... dang... it looked so good. It just went a little longer.
@Remko76 commented on GitHub (Feb 2, 2024):
I used that fps because that is what the stream offered. Must have overread that 5 is the recommendation. Changed it now and that also makes a quite a…
@shad0wca7 commented on GitHub (Feb 4, 2024):
I've just done all of that - also deleted the recordings and database - so the only thing left was the config.yml
Same behavior for me unfortunately..
@tkffaul commented on GitHub (Feb 2, 2024):
Thats it! I updated the driver and it looks rock solid now! Hope this helps someone else! I'll just push them all up since this looks good!
@NickM-27 commented on GitHub (Feb 2, 2024):
actually wait, why is it running detect at 15 fps? in most cases that is not needed and 5 is the recommended default
@NickM-27 commented on GitHub (Feb 2, 2024):
interesting, I am not sure why there would be a difference