If you want to try something that might make your GPU work again, below are the steps.
PaddlePaddle was updated to v2.5.2 (up from 2.5.1) and we've not been able to get GPU working on Windows in the latest. We'll keep trying. I'm sure it's a trivial change. It's just which trivial change...
With the release of CodeProject.AI v2.3.x I need to make some changes to RKNN license plate reader for it to work.
What you can do is clone a camera and just use this camera for face detection. Then in the clone camera's AI settings add the Override server IP and port for the CodeProject.AI server running the face module
Interesting. This screenshot shows Coral version 1.62. I'm running 1.51 and CP 2.2.4
Below is some more details on the module version release notes
/ Which server version is compatible with each version of this module.
"ModuleReleases": [
{ "ModuleVersion": "1.0", "ServerVersionRange": [ "2.1", "2.1.12" ], "ReleaseDate": "2023-07-11" },
{ "ModuleVersion": "1.1", "ServerVersionRange": [ "2.1", "2.1.12" ], "ReleaseDate": "2023-07-12" },
{ "ModuleVersion": "1.2", "ServerVersionRange": [ "2.1", "2.1.12" ], "ReleaseDate": "2023-07-12" },
{ "ModuleVersion": "1.3", "ServerVersionRange": [ "2.1", "2.1.12" ], "ReleaseDate": "2023-08-11", "ReleaseNotes": "installer corrections, macOS/Ubuntu support improved" },
{ "ModuleVersion": "1.4", "ServerVersionRange": [ "2.2", "2.2.4" ], "ReleaseDate": "2023-09-09", "ReleaseNotes": "Updated installer, updated TF-lite runtime" },
{ "ModuleVersion": "1.5", "ServerVersionRange": [ "2.2", "2.2.4" ], "ReleaseDate": "2023-09-16", "ReleaseNotes": "Updates to help Blue Iris users" },
{ "ModuleVersion": "1.5.1", "ServerVersionRange": [ "2.2", "2.2.4" ], "ReleaseDate": "2023-09-17", "ReleaseNotes": "Better checks for admin rights when installing" },
{ "ModuleVersion": "1.6", "ServerVersionRange": [ "2.3.0", "2.3.0" ], "ReleaseDate": "2023-10-01", "ReleaseNotes": "Updated to match new installer SDK." },
{ "ModuleVersion": "1.6.1", "ServerVersionRange": [ "2.3.1", "" ], "ReleaseDate": "2023-10-10", "ReleaseNotes": "Updated to match new installer SDK." },
{ "ModuleVersion": "1.6.2", "ServerVersionRange": [ "2.3.1", "" ], "ReleaseDate": "2023-10-28", "ReleaseNotes": "Improvements for situations where installer does not have admin rights." },
{ "ModuleVersion": "1.6.3", "ServerVersionRange": [ "2.3.1", "" ], "ReleaseDate": "2023-10-28", "ReleaseNotes": "Corrections to requirements.txt for Raspberry Pi / Orange Pi." }@wpiman The License Plate Reader for the Orange Pi 5 should show up in the Install Modules list
Yeah, that was my question as well. For now, I'm sticking with Coral because I have to leave this project (my parents house) and head out of here, but if there are no improvements to speak of, when I come back in Feb, I'll likely buy a GTX 1650 and test it out that way for comparison. My performance gains are there (~120 ms Medium model vs over double on CPU only) but the real issue for me was system instability with a flood of requests, causing it to nearly freeze a few times. That's solved with Coral, but I also tweaked BI Motion & IVS before it can make it to the AI stage. For an A/B comparison without the MD tweaks, Coral handled it without locking the system up, but as you said, legitimate requests got dumped while it was trying to figure out if a human was walking in a windy zone. Next go around, I want to play with this some more. One of my other factors is delayed Push notifications. Although AI might be a minor issue, the root cause is elsewhere since sending alerts right away is still slow & the system triggers it fairly quick.
Just found this on Codeproject boards.
Thanks for this post, set mine up the same way, wasn't using 2.3.4. I was waiting for exactly this before I updated. So far it truly does seem to be working nicely.
Can you post a screenshot of you system info. Version 2.3.4 is working fine with Nvidia GPUs
