CodeProject.AI Version 2.0

[Corvus85]

@Corvus85 @wpiman is correct with the syntax (assuming your unraid container was previously on the last latest release of CPAI).
Docker <--- this shows the exact image names as they get replaced if you want to pull an older one.
Here is where in Unraid you would set it.
screen cap 1, normal mouse click on the CPAI icon in the docker tab. This brings up the menu as you see in screen cap 1. Select edit.
This brings up screen cap 2.
In the noted field (ignore all the other fields that may show) this is where/how you tell docker to pull the image from.

I am using "Plex" as my example in screen cap 1 & 2.
In the final screen cap I show you exactly how it should look for it to pull the last version (with GPU & CUDA).

codeproject/ai-server:gpu-2.1.9

Then you have to scroll to the bottom and hit "apply". The container will update / create a new one & remove the current one automatically.

Hope that helps and solves your issue. Otherwise you will have some more digging to do.View attachment 169807View attachment 169808View attachment 169809

Thanks so much for the detailed response. This is exactly what I did to revert to 2.1.9, and as predicted, it's all working flawlessly.
My question is, how will I ever be able to update to a newer version, given that they'll all likely contain the same bug? The developer on the CP.ai forums hasn't really shown much interest in finding out the source of the problem.

 

[slidermike]

I see some options @Corvus85
#1 Stay on the current version. "if it works do not mess with it" is a very common thought process within the ipcam community.
#2 Try the updated non-cuda gpu version to see if that works for you. (you have 2.1.9 gpu which works to use)
#3 Try the updated gpu version again and see if you just had a bad install the last time. (you have 2.1.9 gpu which works to use)
#4 Monitor CPAI for updates and try new releases as they come out. Falling back each time if they fail.

It is unlikely you will forever and always have to stay on 2.1.9
Be mindful of any/all updates. Read the "whats new" and determine if that patch is for you. As you now know, not everything new and shiny is good.

 

[Pentagano]

Random question here.

What intel cpu would match the inference speeds of a gpu like the gtx970 - 45-75ms?

 

[kbgator]

How do we improve the animal labels? Codeproject, using all models including ipcam-animal, keeps seeing clear raccoon images as cats or dogs.

 

[Pentagano]

How do we improve the animal labels? Codeproject, using all models including ipcam-animal, keeps seeing clear raccoon images as cats or dogs.

hmm I imagine their model of racoons is kind of limited. Ever looked into making your own model with a collection of your racoons?

and if you exclude dog/cat I guess it won't pick up racoon unless you push many images fast it might identify the occasional one

 

[kbgator]

hmm I imagine their model of racoons is kind of limited. Ever looked into making your own model with a collection of your racoons?

and if you exclude dog/cat I guess it won't pick up racoon unless you push many images fast it might identify the occasional one

Thanks. Do you know a link to tutorial on how to build my own model? I can’t seem to find one. Maybe YouTube?

 

[Pentagano]

Thanks. Do you know a link to tutorial on how to build my own model? I can’t seem to find one. Maybe YouTube?

youtube it. I built one about 3 years ago for deepstack, I don't recall really, was a bit fiddly and took 12 hours to build on an i7 7700. There is an amazon cloud platform I believe you can run it on also
Not sure about CP.

 

[wpiman]

Thanks so much for the detailed response. This is exactly what I did to revert to 2.1.9, and as predicted, it's all working flawlessly.
My question is, how will I ever be able to update to a newer version, given that they'll all likely contain the same bug? The developer on the CP.ai forums hasn't really shown much interest in finding out the source of the problem.

I would try eliminating all of the data for code project AI and then installing a clean docker 2.1.10. I had some issues upgrading myself, and eventually trashed my docker directory and started fresh.

Again, I know how to do this in straight linux. Not sure about promox or unraid.

 

[hopalong]

I am currently using BI 5.7.8.1 on a Win10 machine and CodeProject AI 2.0.8 in an unraid container. What are the settings or method to only get AI Confirmed Marked-Up JPGs into a specific folder?

 

[Pentagano]

Been playing around with vmware exsi 7.
Span up a win 10 vm. Installed bi and codeproject etc
Seems to be stable and lightweight.
Plan is to replace my windows server for this.
Also will run truenas on the host
Just hope I can pass through my gtx970.
Had an amd 4100 I passed through just to test but that does not function with codeproject. Showed up in devices

 

[kbgator]

I am currently using BI 5.7.8.1 on a Win10 machine and CodeProject AI 2.0.8 in an unraid container. What are the settings or method to only get AI Confirmed Marked-Up JPGs into a specific folder?

This is the setup I have.
and just select your jpeg storage folder

 

[hopalong]

This is the setup I have.
and just select your jpeg storage folder

See my separate thread here for the issue I am having: Confirmed AI Marked-Up images only

 

[CCTVCam]

@MikeLud1

How do you install Coral Dual Edge TPU's on a Windows platform?

I've seen your post on AI the easy way. But the download seems to be another full AI rather than a Coral module (which should be available from the web menu anyway in CPAI shouldn't it?), and also I know Coral said you needed to add 2 peices of software to Windows to make it work, but I wasn't sure if these were needed with CPAI or whether CPAI's COral module contained these and in any event I can't seem to find out what they were called!

Can you assist? Thanks.

 

[Pentagano]

What a couple of days!! Migrating my windows server to proxmox running as a vm and gpu passthrough!! Went through 4 iterations of windows vm's at least.
Blew gaskets trying to get the gpu working gtx970. Used so many guides and found only one worked for my system.
Was that grub line. So many options out there

This is the definitive one that worked for me.

GRUB_CMDLINE_LINUX_DEFAULT="quiet amd_iommu=on pcie_acs_override=downstream,multifunction video=efifbff video=vesaff vfio-pci.ids=10de:13bb,10de:0fb vfio_iommu_type1.allow_unsafe_interrupts=1 kvm.ignore_msrs=1 modprobe.blacklist=radeon,nouveau,nvidia,nvidiafb,nvidia-gpu"


Had truenas scale to run as a vm also and passthrough 4 sata drives without a HBA SAS card. 1st time lost all data on the raidz2 and spent 19 hours rebuilding it from a back up

Anyways most importantly Blueiris running now on the VM and gpu working on Yolov5.NET BUT not on 6.2 yet oddly. Stuck on CPU that one. Whereas it worked perfectly on the windows before.
Will figure it out eventually. Something amiss

 

[tedrpi]

Looks like 2.1.11 is out, anyone upgrade? Any issues?

CodeProject.AI Server - CodeProject.AI Server v2.5.3

A Guide to using and developing with CodeProject.AI Server

www.codeproject.com

 

[Pentagano]

Looks like 2.1.11 is out, anyone upgrade? Any issues?

CodeProject.AI Server - CodeProject.AI Server v2.5.3

A Guide to using and developing with CodeProject.AI Server

www.codeproject.com

well my gpu stopped working with YOLOv5 6.2.. but might be related to the move to a windows vm and passthrough. Not sure yet. v5 .NET is working on gpu luckily

 

[gwithers]

2.1.11 works like 2.1.9 did and fixed an issue with the docker container for 2.1.10 that was crashing YOLOv5 6.2. So far, a couple days running the docker container and all is well. Processed 59k images with no issues.

 

[Pentagano]

Any ideas why 6.2 will not work with my gpu gtx970 now?
Works fine on the .NET. This is since I moved to a virtualised windows environment.
Am I missing some files/programs for the cuda to work?

GPU passthrough is fine

 

[slidermike]

I have a similar result with my intel based bare-bone Windows 11 build @Pentagano .
.net lets me select GPU (built in intel gpu) but 6.2 bounces back to CPU only.

 

[allgrind]

Yes, it defaults to medium, but I manually set it to small. Medium results in very high latency, worse than even an old p400 video card, so it defeats the purpose.

If the coral can't reliably be used at "small," there isn't much purpose for it.

The alternative is that the default trained models are just very poor compared to yolo -- resulting in inaccuracies for cctv

Hey, recently I also wondered where the real difference is... But strictly speaking, we're comparing two different models.

The Coral module is filled with unnecessary items for us, so it probably takes longer. Mike's optimized modules really focus on the essentials and naturally optimize the processing times.

With Mike's modules, my CPU takes about 300ms per iteration, while my Coral with the standard modules takes about 200 to 250ms.

If customized modules for the Coral come out, the times should significantly improve.

What do others think about this?