Tayschrenn
Young grasshopper
So, here's a screen shot of the event in question:
Here's a freeze fame from the export:
You can clearly see that it detects motion - but there's NO alert generated, and thus no Ai analysis run by CPAI.
When I go back I can't select the test/tune to see if AI detection would even go against it which doesn't help - so how do I troubleshoot this. I'm only using CPU Ai in a docker container so I know it's not 'fast' fast, but my only other option currently is to run a laptop on the same network with a Quadro GPU to do the detection; I've tested CPAI on it and it's for sure faster, but I feel like the latency involved with sending the image over makes it less ideal - although so long as BlueIris is saving the alert locally before sending it, I suppose that's less an issue?
Really that's my only initial guess on why it was a 'miss' - because the CPU was too busy to detect the motion perhaps. Though that doesn't explain why the Export shows the motion detection =(
Am I missing something?
The camera settings:
It's worth noting I do have 'Fox' on the CPAI call out - but apparently it... wouldn't have really caught it even if it DID alert - but the fact is it didn't save an alert at all.
Although apparently when it's running, it looks like a squirrel... I guess that's not saying much as it also frequently says my tree leaves look like cows and humans day to day.
Man I will say my detection times are getting worse - today it's looking like they're going from 400ms-1200ms
- I use YOLOv5 6.2 because .NET - despite being 'CPU' designed seems to be slower and less accurate; I run the large model. My CPAI docker container has 2vCPU and can run up to 10GB RAM - the BI machine has an i5-9500 and 32GB of memory total - but I only have 4 cameras so my ram rarely goes over 17GB used total, and CPU average seems 15%-50% but spikes to 100% a fair amount.
Here's a freeze fame from the export:
You can clearly see that it detects motion - but there's NO alert generated, and thus no Ai analysis run by CPAI.
When I go back I can't select the test/tune to see if AI detection would even go against it which doesn't help - so how do I troubleshoot this. I'm only using CPU Ai in a docker container so I know it's not 'fast' fast, but my only other option currently is to run a laptop on the same network with a Quadro GPU to do the detection; I've tested CPAI on it and it's for sure faster, but I feel like the latency involved with sending the image over makes it less ideal - although so long as BlueIris is saving the alert locally before sending it, I suppose that's less an issue?
Really that's my only initial guess on why it was a 'miss' - because the CPU was too busy to detect the motion perhaps. Though that doesn't explain why the Export shows the motion detection =(
Am I missing something?
The camera settings:
It's worth noting I do have 'Fox' on the CPAI call out - but apparently it... wouldn't have really caught it even if it DID alert - but the fact is it didn't save an alert at all.
Although apparently when it's running, it looks like a squirrel... I guess that's not saying much as it also frequently says my tree leaves look like cows and humans day to day.
Man I will say my detection times are getting worse - today it's looking like they're going from 400ms-1200ms
- I use YOLOv5 6.2 because .NET - despite being 'CPU' designed seems to be slower and less accurate; I run the large model. My CPAI docker container has 2vCPU and can run up to 10GB RAM - the BI machine has an i5-9500 and 32GB of memory total - but I only have 4 cameras so my ram rarely goes over 17GB used total, and CPU average seems 15%-50% but spikes to 100% a fair amount.