CodeProject.AI Version 2.5

Attached is a new YOLOv8 license-plate model. I did away with the DayPlate and NightPlate labels, now the new label is just Plate. To use the model just save the model in the below folder and then restart Blue Iris service.

1708913733853.png
 

Attachments

Last edited:
Attached is a new YOLOv8 license-plate model. I did away with the DayPlate and NightPlate labels, now the new label is just Plate. To use the model just save the model in the below folder and then restart Blue Iris service.

Mike: is Yolov8 ready for prime time.?.. can it be installed in a win10 system using just the cpu without a graphics card (now using Yolov5.net )? If so, does it offer advantages in speed or accuracy? And if yes to all that, are you re-doing the ipcam-combined model for it? Thanks...
 
Attached is a new YOLOv8 license-plate model. I did away with the DayPlate and NightPlate labels, now the new label is just Plate. To use the model just save the model in the below folder and then restart Blue Iris service.

View attachment 187755
Should the naming of these models identify with which YOLOversion since the old models don't work with YOLOv8?

ex.
ipcam-general-yolov8.pt
license-plate-yolov8.pt
 
Should the naming of these models identify with which YOLOversion since the old models don't work with YOLOv8?

ex.
ipcam-general-yolov8.pt
license-plate-yolov8.pt

@David L If you name the models as suggested, won't you need to go into every camera AI screen and update custom model config when you switch from v8 version back to v5, and vice versa? I'm not sure this is a good thing. Increases the chances I will have typos or misconfigure, given the #of cameras where I do AI
 
Last edited:
  • Like
Reactions: David L
@David L If you name the models as suggested, won't you need to go into every camera AI screen and update custom model config when you switch from v8 version back to v5, and vice versa? I'm not sure this is a good thing. Increases the chances I will have typos or misconfigure, given the #of cameras where I do AI
I gotcha, just thinking of these files out there, how do you you know which is for what YOLO version unless the filenames are different.
 
Anyone try the 2.5.5 update? I tired and couldn't get the status page (localhost:32168) to load, reverting to 2.5.4 now.
 
It looks like there is an issue with 2.5.5, DO NOT UPDATE YET.
Too late on the above note for me. I updated it this morning, and now the server service wont start, even after a few tries and restarts. I will try reverting to the 2.5.1 version that i have downloaded next.

[EDIT]: It looks like the reversion back to 2.5.1 is working *server service started right up). So far noting is being processed (YOLOv5 6.2) , but expect to figure out why soon.
[EDIT 2]: restarting the CPAI service seemed to do the trick, I am now on YOLOv5 .NET, and have enabled GPU to use my Intel GPU, (which does not seem to improve speed, so maybe i should test it both ways?)
 
Last edited:
I gave up on 2.5.4, no matter what I tried it wouldn't go into GPU (CUDA) mode (at least on the dashboard displayed info). Was about to go 2.5.5 then saw this lol Went back to 2.5.1 and it's all ok again so I will give 2.5.5 a go once they get that working.
 
It looks like there is an issue with 2.5.5, DO NOT UPDATE YET.

Yep, I couldn't get 2.5.5 to load either after install. Switched back to 2.5.4. I like the mesh feature as I can update my secondary system and make sure its stable before updating the CPAI that's running on my BI box.
 
Mike: is Yolov8 ready for prime time.?.. can it be installed in a win10 system using just the cpu without a graphics card (now using Yolov5.net )? If so, does it offer advantages in speed or accuracy? And if yes to all that, are you re-doing the "ipcam-combined" model for it? Thanks...
Mike: re asking, with the added question of using Yolov8 in a win 11 box with a RTX 3070 GPU (as a mesh).
 
Mike: re asking, with the added question of using Yolov8 in a win 11 box with a RTX 3070 GPU (as a mesh).
YOLOv8 does not have all the custom models yet so if you are using the default models or ipcam-general then give it a try. For your GPU the best Object Detection module would be Object Detection (YOLOv5 .NET). On my 3060 I am get the below speeds.

1709314917745.png
 
  • Like
Reactions: rdxny
YOLOv8 does not have all the custom models yet so if you are using the default models or ipcam-general then give it a try. For your GPU the best Object Detection module would be Object Detection (YOLOv5 .NET). On my 3060 I am get the below speeds.

View attachment 188229
Thanks Mike... I'm using Yolov5.net on the cpu only BI machine, and was using Yolov5 6.2 for the Win11 3070gpu pc in the mesh, but I'll try moving that one back to Yolov5.net. I look for deer (to dissuade them!), so I don't think IPcam-general works for that, so I will avoid Yolov8 until you do the custom models like ipcam-combined for v8. Many thanks for your hard work!
 
What's the best way to run codeproject AI so it only has "people" models... that is all I care about, i dont care about cars, animals, etc. I feel that if I can eliminate those models which arent required it should speed things up quite a bit.
 
What's the best way to run codeproject AI so it only has "people" models... that is all I care about, i dont care about cars, animals, etc. I feel that if I can eliminate those models which arent required it should speed things up quite a bit.

Just run the ipcam-general one and tell it to only IDENTIFY person. All that model has is person and vehicle. It is a fast model.