One of the main reasons why I am still on Deepstack. I know many run CodeProject just fine, but we never saw as many issues with Deepstack as we do with Code Project.
CodeProject.AI Version 2.0
- Thread starter MikeLud1
- Start date
When I was testing CodeProject I noticed that it would throw that error anytime I disconnected the computer from the internet, which would be similar to their website is down and installing the Microsoft loopback adapter fixed it:
right click on window start menu icon and select Device manager. Device manager window will immediately open (or you may use any other way how to open device manager window)
click on Action, and select Add legacy hardware
click Next on welcome screen
choose "Install the hardware that i manually select from a list" and click on Next
scroll down and select Network adapters from offered common hardware types and click on Next
select Microsoft as the manufacturer, and then select Microsoft KM-TEST Loopback adapter card model, click on Next
click on Next
click on Finish
Now that annoying "Unable to contact AI server" error is gone!
CP/AI has been solid here for some time.
Apologies, this is oranges/apples but maybe of some use.
1. In my SOHO, I am running BI 5.8.0.0 in a W10 virtual machine, this is a test lab with 3 cameras. Host is ESXi 7.03
2. I am running CPAI 2.2.4 in a Debian 11 Linux in a VM on the same ESXi system with an ancient NVIDIA K620 via passthrough. Make sure this VM has at least 2 cores, AI doesn't run properly with 1 core (in my experiences early on)
3. Installed cuDNN via tar and compile (not sure this is necessary)
4. Installed the cuda toolkit (also not sure).
5. Installed nvidia-docker2
6. I am using NVIDIA docker. (sudo nvidia-docker run --gpus all -p 32168:32168 codeproject/ai-server:gpu-2.2.4).
7. Made sure I could run some stuff through CP/AI on the Linux machine CP Explorer before even trying BI. Made sure it showed GPU on object detection (mine shows "YOLOv5 6.2 started GPU(CUDA)".
8. My detection times in BI are nothing to write home about but under 100msec.
Anytime I change something in either system, I reboot both VM's.
Our production site has 13 cameras, only 3 are using AI but results similar to above. Also has Linux system with Docker image. Will increase when we upgrade the video card.
Edit: have no hair to offer.
fenderman
Staff member
- Joined
- Mar 9, 2014
- Messages
- 36,905
- Reaction score
- 21,278
This is only because cpai is pushing lots of updates with changes. Deepstack had two versions after it was implemented with BI. Then they did nothing. Cpai 2.0.8 works perfectly across many machines I have. There are also many more users of cpai and it offers a wide variety of options not available in DS.
How did you set up directML?
parsapolis
n3wb
Just wanted to share that I switched to the mini pci-e coral yesterday and my inference speed has gone from ~200ms average on the usb coral with medium model size down to 40ms average. Pretty impressive. I'm on the latest Codeproject and BI.
For comparison, my GPU inference speeds are 150ms average on a gtx1660ti with 6gb ram.
For comparison, my GPU inference speeds are 150ms average on a gtx1660ti with 6gb ram.
Impressive results. Keep us posted in a week or so to let us know how the overall stability/experience is going.
parsapolis
n3wb
parsapolis
n3wb
One interesting difference is the custom model. When using the USB coral that field in BI was blank but with the mini PCI-e coral it shows MobileNetSSD as a custom model.
Last edited:
Interesting...did you find any additional info on this MobileNetSSD model? What objects it can use (person, car, animal, etc...) and whether it is optimized for night use?
Edit: Perhaps more importantly, are your comparison numbers generated from the same (like for like) models? And are the numbers from commonly used models (ipcam-combined, ipcam-general, ipcam-dark, etc...)?
Last edited:
parsapolis
n3wb
Haven't found much only that it's optimized for mobile devices. Maybe someone else can chime in that is familiar with it.
parsapolis
n3wb
Alright looks like the MobileNetSSD model can handle 90 objects. Based on coral.ai it uses the COCO dataset
github.com
What Object Categories / Labels Are In COCO Dataset?
One important element of deep learning and machine learning at large is dataset. A good dataset will contribute to a model with good precision and recall. In the realm of object detection in images…
tech.amikelive.com
coco-labels/coco-labels-paper.txt at master · amikelive/coco-labels
The labels for object categories in COCO dataset. Contribute to amikelive/coco-labels development by creating an account on GitHub.
github.com
Attachments
Last edited:
MikeLud1
IPCT Contributor
It detects the same objects as YOLOv5Alright looks like the MobileNetSSD model can handle 90 objects. Based on coral.ai it uses the COCO dataset
What Object Categories / Labels Are In COCO Dataset?
One important element of deep learning and machine learning at large is dataset. A good dataset will contribute to a model with good precision and recall. In the realm of object detection in images…tech.amikelive.com
coco-labels/coco-labels-paper.txt at master · amikelive/coco-labels
The labels for object categories in COCO dataset. Contribute to amikelive/coco-labels development by creating an account on GitHub.
xm4rcell0x
n3wb
Hi Mike! i'd want to know how to customize AI camera settings to minimize resources waste. I have a ptz camera, that i want to recognize objects, faces and license-plates. In the general AI settings i checked "default objects,custom model, faces, alpr for plates", while in camera AI I have these settings
what if i add : "alpr and faces" in custom models ? Is it necessary ?
In the other cameras that i don't want to do license-plate and faces recognition i have to specify "faces:0,license-plate:0,alpr:0" in custom model or i can simply uncheck "save unknown face to" (in order to disable face recognition) and leave ipcam-general as a custom model ?
Thank you so much.
Hey,
Ive got one of my cameras set up to only confirm people. I noticed when I get deliveries the camera picks up people as they leave my house. How can I get the initial trigger (vehicle) to continue analyzing and pick up the person exiting the vehicle? Would the issue be in the AI option for leading motion and or my Break time under triggers.
Thanks
Ive got one of my cameras set up to only confirm people. I noticed when I get deliveries the camera picks up people as they leave my house. How can I get the initial trigger (vehicle) to continue analyzing and pick up the person exiting the vehicle? Would the issue be in the AI option for leading motion and or my Break time under triggers.
Thanks
Attachments
about face detection
how it works?
if i create a profile with multiple faces on the server from the browser the camera stop the checks and stop to add the unknown faces in my folder
i don't see any alert made with the new face added
how i can add a face and still get unknown faces?
how it works?
if i create a profile with multiple faces on the server from the browser the camera stop the checks and stop to add the unknown faces in my folder
i don't see any alert made with the new face added
how i can add a face and still get unknown faces?