Improving AI Confirmation Time

[clennys]

It's taking me about 15 seconds to go from motion triggered to AI confirmation according to the logs. It used to only take a few seconds. The only thing I can think of is that I've moved my BI installation from its own baremetal PC into a Windows VM on Unraid. My CPU % usage is high but not pegged. I am using CodeProject AI in a docker container using CUDA/GPU.

I found a similar post where the guy got his AI confirmation from 20 seconds to a few seconds by removing the default object detection but I am having trouble with that step. Attached are some of my configs. I have since unchecked "use main stream". This is what my logs look like:

21:38:34:Response rec'd from Object Detection (YOLOv8) command 'custom' (...7e4d14)
21:38:34:Response rec'd from Object Detection (YOLOv8) command 'custom' (...5c28a1)
21:38:35:Response rec'd from Object Detection (YOLOv8) command 'custom' (...1467dc)
21:38:35:Response rec'd from Object Detection (YOLOv8) command 'custom' (...c9d1cf)
21:38:35:Response rec'd from Object Detection (YOLOv8) command 'custom' (...16d82f)
21:38:36:Response rec'd from Object Detection (YOLOv8) command 'custom' (...bb1159)
21:38:36:Response rec'd from Object Detection (YOLOv8) command 'custom' (...9474dc)
21:38:36:Response rec'd from Object Detection (YOLOv8) command 'custom' (...2798bc)
21:38:37:Response rec'd from Object Detection (YOLOv8) command 'custom' (...e4083f)
21:38:37:Response rec'd from Object Detection (YOLOv8) command 'custom' (...cbb997)

When I revert it back to my original settings it looks like this, which is to enable default objects detection and disable custom models:

21:49:03:Response rec'd from Object Detection (YOLOv8) command 'detect' (...5d7ba3) ['No objects found'] took 31ms
21:49:11:Response rec'd from Object Detection (YOLOv8) command 'detect' (...7076e9) ['Found chair'] took 28ms
21:49:22:Response rec'd from Object Detection (YOLOv8) command 'detect' (...18a2dd) ['No objects found'] took 28ms
21:50:03:Response rec'd from Object Detection (YOLOv8) command 'detect' (...cc6550) ['Found bench, chair, chair...'] took 69ms
21:50:06:Response rec'd from Object Detection (YOLOv8) command 'detect' (...f69d90) ['Found car, surfboard'] took 55ms
21:50:08:Response rec'd from Object Detection (YOLOv8) command 'detect' (...82c59a) ['No objects found'] took 34ms
21:50:10:Response rec'd from Object Detection (YOLOv8) command 'detect' (...e1097c) ['Found car, car, car'] took 86ms
21:50:11:Response rec'd from Object Detection (YOLOv8) command 'detect' (...48e2be) ['Found car, surfboard'] took 66ms
21:50:22:Response rec'd from Object Detection (YOLOv8) command 'detect' (...d09bc6) ['Found refrigerator'] took 34ms
21:50:28:Response rec'd from Object Detection (YOLOv8) command 'detect' (...928541) ['Found car, car, airplane'] took 87ms
21:50:36:Response rec'd from Object Detection (YOLOv8) command 'detect' (...d5cdae) ['Found person, car'] took 67ms
21:50:46:Response rec'd from Object Detection (YOLOv8) command 'detect' (...be0639) ['Found car'] took 56ms
21:51:05:Response rec'd from Object Detection (YOLOv8) command 'detect' (...017767) ['Found car, surfboard'] took 109ms
21:51:10:Response rec'd from Object Detection (YOLOv8) command 'detect' (...ea482f) ['Found chair'] took 29ms

 

[clennys]

After looking at some .dat files there is something I've noticed. When the AI has to evaluate for a person at all. The confirmation time is 15+ seconds. When there's only cars, it's fast. I've gone through at least 30 alerts now and it's true 100% of the time. What is weird is that if you look under 'Objects' it only took 289ms to identify person but the confirmation time is still 15353ms. Excuse the screenshots. I am using remote desktop and didn't know how to take a screenshot.