Is it me but do updates seem to impact Codeproject

H

heyho

n3wb
Jun 10, 2024
23
6
England
I am getting increasingly frustrated by Blue iris. My first complaint is that depsite paying good money for the product the company seem incapable of keeping their documention up to date. Luckily this site helps a lot.

But my main concern is recently updates seem to casue more issues than they fix. I know Codeproject is a product that seems to not be developed further. I have been using Yolov8 for some time now and it works in conjunction with Blue iris. but not foir the first time since an update the system appears to check for triggers but the actual categorisation doesnt work, hence no alerts. Only real fix is to back rev BI and also totally reinstall Codeproject.

Are BI gearing for version 6 and no quality checking release updates as stringently?
 
heyho said:
I am getting increasingly frustrated by Blue iris. My first complaint is that depsite paying good money for the product the company seem incapable of keeping their documention up to date. Luckily this site helps a lot.

But my main concern is recently updates seem to casue more issues than they fix. I know Codeproject is a product that seems to not be developed further. I have been using Yolov8 for some time now and it works in conjunction with Blue iris. but not foir the first time since an update the system appears to check for triggers but the actual categorisation doesnt work, hence no alerts. Only real fix is to back rev BI and also totally reinstall Codeproject.

Are BI gearing for version 6 and no quality checking release updates as stringently?
Click to expand...
For a start, the frequent BI updates are beta/test not stable so if you want to avoid issues then stick to the stable and I find the change log and help file more than capable for all my needs. Secondly you mention issues and CP.AI but not what those issues are, so is this just a moan or do you want some help?

FYI - I've been running CP.AI for years now (moved from Deepstack) I also run the bleeding edge versions of both with the exception of the odd bug on the bleeding edge version that Ken often fixes within a few hours I don't have any issues, and I've got a fairly complex setup with phones. MQTT, AI, push etc etc
 
  • Like
Reactions: joegr, heyho, anijet and 1 other person
CrazyAsYou said:
For a start, the frequent BI updates are beta/test not stable so if you want to avoid issues then stick to the stable and I find the change log and help file more than capable for all my needs. Secondly you mention issues and CP.AI but not what those issues are, so is this just a moan or do you want some help?

FYI - I've been running CP.AI for years now (moved from Deepstack) I also run the bleeding edge versions of both with the exception of the odd bug on the bleeding edge version that Ken often fixes within a few hours I don't have any issues, and I've got a fairly complex setup with phones. MQTT, AI, push etc etc
Click to expand...
Not a maon (well ok the poor documentation is) but I was trying to ascertain if others are experiencing this before i take further 'drastic' measures, well maybe reinstall Windows 11 and start again. Seems like it might be me rather than a general issue. By the way my backround is IT support (40+ years) and i used to head a team that one of their functions was to pre test those dreaded microsoft updates before general release (20,000 staff!).

I have used CP.AI for around 2 years now running Yolov5 6.2 upto about 6 months ago. I would regularly apply Microsoft patches, graphics driver updates (Nvidia) plus BI updates on a regulalr basis with no issues. then one day i started getting a lot of errors in the CP.AI logs and the detection within BI stopped. Tried several things including a full reinstall of CP.AI but to no avail. Tbh I'd recently (not straight away) updates microsoft and nvidia drivers.

Anyway I ended up with Yolov8 which once again worked fine until a couple of weeks ago when, after a BI update, there seemed to be a loss of communication between an alert being recognised within BI and CP.AI assessing it for classification. tried all sorts including full reinstall of CP.AI and even reverting back to Yolo5 6.2 (that still showed up errors in log). I thought I had resolved the issues with Yolov8 only to find everything got classified as a person. Before a car was a car with no issues.

Now you guys have answered my 'is it me' question I am planning to reinstall Windows 11 from scratch and then BI and CP.AI (might even try Yolov11). At the moment I've uninstalled CP.AI and its detecting fine (obvioulsy withoutthe AI classifiation).

Theres a fair bit of reliance - Nvidia graphics, CUDA, .NET??? etc so a full rebuild might start to answer a few question.

Thanks for your response
 
Is the issue BI talking to CI.AP or do you have an actual issue with CP.AI? Easy way to tell is access to the CP.AI local web GUI and see what's running and perform some tests. FYI I had Nvidia/CUDA working very well for years and recently after a huge amount of testing for performance I moved over to Direct ML. on my rig not only is this performing well but also means I can skip a huge python install and all the faff with CUDA/cuDNN - BTW on Windows 11 before going to the extreme of a fresh install check and double-check the basics such as Windows firewall for example.
 
Is the issue BI talking to CI.AP or do you have an actual issue with CP.AI? Easy way to tell is access to the CP.AI local web GUI and see what's running and perform some tests. FYI I had Nvidia/CUDA working very well for years and recently after a huge amount of testing for performance I moved over to Direct ML. on my rig not only is this performing well but also means I can skip a huge python install and all the faff with CUDA/cuDNN - BTW on Windows 11 before going to the extreme of a fresh install check and double-check the basics such as Windows firewall for example.
 
CrazyAsYou said:
Is the issue BI talking to CI.AP or do you have an actual issue with CP.AI? Easy way to tell is access to the CP.AI local web GUI and see what's running and perform some tests. FYI I had Nvidia/CUDA working very well for years and recently after a huge amount of testing for performance I moved over to Direct ML. on my rig not only is this performing well but also means I can skip a huge python install and all the faff with CUDA/cuDNN - BTW on Windows 11 before going to the extreme of a fresh install check and double-check the basics such as Windows firewall for example.
Click to expand...
Decided not to be lazy and do a fresh Windows install so I:

- fully uninstalled CodeProject and all its directories
- uninstalled CUDA, CuDNN and NVIDIA drivers
- did a CCleaner type clean up
- reinstalled NVIDIA drivers only
- reinstalled CodeProject
- manually installed YoloV8 (GitHub download and command line prompt)

Not a lot goes on in my street but it has correctly identified and classified a car.

I did have my suspicions over CUDA/CuDNN as I had updated them a while back (but it still worked for several weeks after). Codeproject is showing GPU (Cuda) successfully).

Anyway time will tell.

Out of interest what version of Yolo are you using and have you instrinctions to use DirectML rather than Cuda for it. I'd prefer to use that. I did have a little play with YoloV11 which used DirectML but was still having issues.
 
As part of my CUDA/DirectML testing I also included Yolo8 but didn't see any improvement in detection(using Standard Large model), but I did see increased processing and large increases in GPU/memory consumption - I much prefer the smaller footprint of .NET with DIrectML. I want superfast lightweight processing as BI and CP.AI handoff to my home automations to perform other actions I want them within 1-2seconds of camera triggers, for me the Min(ms) with decent accuracy is my main goal. The Max/AVG in the below table can be ignored, most of that is when something is not found.

1762793518894.png

FYI - DirectML kicks in when you enable GPU via the little settings cog/gear for Object Detection (YOLOv5 .NET)

1762793687483.png
 
CrazyAsYou said:
As part of my CUDA/DirectML testing I also included Yolo8 but didn't see any improvement in detection(using Standard Large model), but I did see increased processing and large increases in GPU/memory consumption - I much prefer the smaller footprint of .NET with DIrectML. I want superfast lightweight processing as BI and CP.AI handoff to my home automations to perform other actions I want them within 1-2seconds of camera triggers, for me the Min(ms) with decent accuracy is my main goal. The Max/AVG in the below table can be ignored, most of that is when something is not found.

View attachment 232206

FYI - DirectML kicks in when you enable GPU via the little settings cog/gear for Object Detection (YOLOv5 .NET)

View attachment 232207
Click to expand...
Classic example of what frustrates me about CP.AI. 5 hours worth of Yolov8 and a car is a car, a person is a person etc etc.

So i installed Yolov11 and switched Yolov8 off. It went to Direct XL but i wasnt happy with the classification so i went back to Yolov8. Straight away a car is a bus (95%) and also a person (94%). Have tried the full uninstall route I did before but to no avail.
 
Are you running a supercomputer at home? :) - I'm no Yolo expert, but I thought Yolo11 was the reserve of smart cars and robots that needed super high accuracy with dedicated AI HW. I find Yolo5L perfect for my needs, maybe I might go Yolo8M, but that would be it for home CCTV automation.
 
Th
CrazyAsYou said:
Are you running a supercomputer at home? :) - I'm no Yolo expert, but I thought Yolo11 was the reserve of smart cars and robots that needed super high accuracy with dedicated AI HW. I find Yolo5L perfect for my needs, maybe I might go Yolo8M, but that would be it for home CCTV automation.
Click to expand...
The only reason I went to YoloV8 was because 5 6.2 started coming back with loads of errors ( I do suspect it might have coincided with and upgrade to CudNN. Today I wiped my computer, reinstalled Windows, Nvidia drivers, CP.AI and BI from scratch and went back to Yolov5 6.2. Everything appears fine. I shall be leaving it alone!

At the moment I am using 'Default object detection'. Are you just using this or combing it with or replacing it with custom models. And if so which ones are you using - actionnetv2, ipcam-animal, ipcam-combined, ipcam-dark, ipcam-general (I dont use licence plat recognition). Ive never got my head round which ones to use tbh.
 
You must log in or register to reply here.
Top Bottom