After spending much time testing my gpu with CPAI on all the settings (and camera settings) I have found that using the main stream for AI analysis (sub recording + triggers) has resulted in a far higher accurate confidence level especially in the dark.
The inference times are not that much longer either.
I have the snapshots set at 333ms intervals now (before was at 100- 250)
Inference speeds are about 90ms-150ms.
Fine for me considering the improvement in object detection.
10s trigger end after 20.
20 + real images @333ms
New trigger and retriggers selected also.
Zones
I guess it all depends on your set up, environment and hardware.
Cheers!
The inference times are not that much longer either.
I have the snapshots set at 333ms intervals now (before was at 100- 250)
Inference speeds are about 90ms-150ms.
Fine for me considering the improvement in object detection.
10s trigger end after 20.
20 + real images @333ms
New trigger and retriggers selected also.
Zones
I guess it all depends on your set up, environment and hardware.
Cheers!