CodeProject.AI Version 2.0

gwithers

Getting the hang of it
Joined
May 18, 2016
Messages
49
Reaction score
38
Easiest way would be to just use one of Mike's custom models that are included with the CPAI install. Specifically use only the "IPcam-general" model which only has "person" and "vehicle" as objects to be detected. I am not sure what a trailing comma will do in your "to confirm" list. Probably nothing but it probably shouldn't be there either. You may also want to increase the confidence needed, at least for the person. It all depend on the field of view but I find 75-80% is a minimum needed for detecting persons. Wider fields of view will require lower confidence setpoints typically though. Appropriate thresholds will remove a lot of low confidence stuff in the 50% range right off the bat. This probably is less to do with the CPAI itself and more to do with the model you are using. You will get different results with different models.
 

Post

Getting the hang of it
Joined
Sep 1, 2016
Messages
56
Reaction score
29
Thanks! I'll give that a try and report back.
 

swilliams76360

Getting the hang of it
Joined
Oct 5, 2022
Messages
29
Reaction score
25
Location
texas
Has anyone tested out the Facial Rec on this new version yet? I am really wanting to use that feature to basically filter out my roommate and I from the alerting to phone. I tried it once before and it was like a 25% chance it detected us correctly. Now that i have finally got my system fully updated and stable with the new power supply, I am willing to get it another go. Question is do I need to have multiple different images of us registered in the system?
 

Post

Getting the hang of it
Joined
Sep 1, 2016
Messages
56
Reaction score
29
So, I'm now using the custom model (ipcam-general) and am still getting alerts with the "dog" as "person" (78% confidence in attached example). I also updated to BI 5.7.0.5. Any other suggestions?

1678726217880.png1678726018165.png1678726117168.png
 

MikeLud1

IPCT Contributor
Joined
Apr 5, 2017
Messages
2,141
Reaction score
4,118
Location
Brooklyn, NY

Attachments

Post

Getting the hang of it
Joined
Sep 1, 2016
Messages
56
Reaction score
29
Mike, I searched my PC for "ipcam-general" and found two files types (PT and ONNX). I'm using the Object Detection YOLOv5.net module so I'm assuming I need the ONNX type (the file you sent is PT)?


1678730180554.png


1678730094001.png
 

MikeLud1

IPCT Contributor
Joined
Apr 5, 2017
Messages
2,141
Reaction score
4,118
Location
Brooklyn, NY

MikeLud1

IPCT Contributor
Joined
Apr 5, 2017
Messages
2,141
Reaction score
4,118
Location
Brooklyn, NY

Attachments

gwithers

Getting the hang of it
Joined
May 18, 2016
Messages
49
Reaction score
38
So, I'm now using the custom model (ipcam-general) and am still getting alerts with the "dog" as "person" (78% confidence in attached example). I also updated to BI 5.7.0.5. Any other suggestions?

View attachment 156727View attachment 156725View attachment 156726
You need to deselect "default object detection". You are still using the standard model with that selected, even with the custom selected and settings for the camera AI that you have.
 

Post

Getting the hang of it
Joined
Sep 1, 2016
Messages
56
Reaction score
29
Well, unfortunately no luck. I installed the new ONNX version, unchecked Default Object Detection and then restarted the PC. Still getting alerts with the dog as "person: 79%". Any other ideas?
 

gwithers

Getting the hang of it
Joined
May 18, 2016
Messages
49
Reaction score
38
Well, unfortunately no luck. I installed the new ONNX version, unchecked Default Object Detection and then restarted the PC. Still getting alerts with the dog as "person: 79%". Any other ideas?
Sorry, I'm out of ideas.
 
Last edited:

hapstabu

Getting the hang of it
Joined
Aug 29, 2020
Messages
59
Reaction score
38
Location
US
Code project AI ver 2.0.8
I had an issue yesterday where Code project stopped analyzing images but kept queing images for quite a while.
Blue Iris kept recording for about 7 more hours until it quit.
This morning, I had no user interface, the machine was hung.
I manually rebooted a couple times to get through Windows update, then went back to Deepstack for now.
Has anyone run Code project Docker on a Raspberry PI?
What is your experience? How does it run? (with cpu, no gpu I assume)
Is it slow?
 

Village Guy

Pulling my weight
Joined
May 6, 2020
Messages
291
Reaction score
161
Location
UK
This is a heads up for anyone using a GPU accelerator with BI.

My BI has been logging quite a few Error 500's and after looking at the A.I server log I can see a lot of connection refused errors. After setting BI to NO hardware acceleration, the system has run error free for 10 hours. It would appear that BI and CodeProject server do not play nicely together. Fortunately my CPU has enough power to work just fine without GPU acceleration in my case.

This is the error I found in the A.I server log.

2023-03-15 05:32:49: Object Detection (YOLOv5 6.2): Queue and Processing Object Detection (YOLOv5 6.2) command 'custom' (#reqid b4636fc0-b7b0-4709-8b9d-f7fb21f84219) took 100ms (command timing) in Object Detection (YOLOv5 6.2)
2023-03-15 05:32:59: ModuleRunner Stop
2023-03-15 05:32:59: Sending shutdown request to python/ObjectDetectionYolo
2023-03-15 05:32:59: Client request 'Quit' in the queue (#reqid af7985ac-67bb-4a17-bd9b-214801cb0587)
2023-03-15 05:33:01: detect_adapter.py: Not using half-precision for the device 'NVIDIA GeForce GTX 1060 6GB'
2023-03-15 05:33:01: detect_adapter.py: [ConnectionRefusedError] : Unable to check the command queue objectdetection_queue. Is the server running, and can you connect to the server?objectdetection_queue: [ConnectionRefusedError] : Unable to check the command queue objectdetection_queue. Is the server running, and can you connect to the server?
2023-03-15 05:33:01: detect_adapter.py: Inference processing will occur on device 'NVIDIA GeForce GTX 1060 6GB'
2023-03-15 05:33:01: detect_adapter.py: Timeout connecting to the server
2023-03-15 05:33:01 [ConnectionRefusedError]: Unable to check the command queue objectdetection_queue. Is the server running, and can you connect to the server?
 
Last edited:

AlphaBlueIris

Getting the hang of it
Joined
Jan 17, 2022
Messages
125
Reaction score
27
Location
Boston
Please I need your help, I can not figure it out no matter what I do. It sees Amazon just fine and it recognize it but the log shows something (icons) that I do not understand. I'm, almost there :) I also disabled Plate but it still shows that it is AI it. Any help would be great.
 

Attachments

AlphaBlueIris

Getting the hang of it
Joined
Jan 17, 2022
Messages
125
Reaction score
27
Location
Boston
Thank you. The delivery works great, for some reason ai is not marking it as delivery, any idea?

Sent from my SM-S906U using Tapatalk
 

truglo

Pulling my weight
Joined
Jun 28, 2017
Messages
275
Reaction score
103
I have doubts it's the cause of delivery not working for you, but you probably should remove all of the "Mark as vehicle" entries since you aren't confirming plates or vehicles on this one.

I vaguely remember seeing those blue/orange double rectangle icons in the ai logs when ai was misbehaving with detecting static objects (especially without alpr:0). I noticed you have that disabled. Maybe enable it to see if it works?
 
Top