5.5.8 - June 13, 2022 - Code Project’s SenseAI Version 1 - See V2 here https://ipcamtalk.com/threads/codeproject-ai-version-2-0.68030/

Hmmm... I installed it on two machines with no issues on either one.
yep...more than likely just one of those "who knows" things. Seems to run fine now. Looking forward to a more streamlined upgrade process though from CP.AI Or better yet, a download version that requires no further downloading of components during the install process.
 
Could you explain that in more detail Mike?
 
I don't get how choosing model size choose which yolo? is there a table of model size and yolo? like medium= 5l or something?

also : is there a difference between yolov5l and yolov5l-cls?

the one I downloaded from git is way smaller than the one installed by ci by default, like 50 vs 90
 
What's the difference between the sizes?

And what's the difference between yolov5l, yolov5m,yolov5n, and yolov5s?
What objects do they each include?

They all have the same objects. The difference is how accurate they are. The bigger the model the better the accuracy.

This comes at a cost of extra CPU/GPU cycles and an increased memory footprint. The extra CPU/GPU cycles will almost mean slightly longer detection times.

The best thing to-do is take a whole bunch of snapshots from your own cameras and run them all via the “Object Detection” in the CP.AI Explorer in the web browser to compare what difference you get for objects detected and the percentage confidence, then pick the smallest yolo model you’re happy with.

In my testing I see great (accurate) results with yolov5l and yolov5m and ok results with yolov5n and yolov5s.

As I have a GTX 970 that handles yolov5l pretty well <80ms results and as it scores about 5%-9% better confidence on my cameras for people and cars than yolov5m I run the larger one(yolov5l) but each persons mileage will vary depending on camera images and system performance.
 
Interesting. How does this compare with the ipcam custom models? Especially when detecting animals?

I can't comment on animals as I only look for people and cars. The larger yolo model is more accurate that ipcam-general but as I said it comes at the cost of speed and system resources. As I recall ipcam-general performed almost the same level (slightly lower confidence) of accuracy as yolov5m but is quicker due to the reduced objects it can detect.
 
Is there a central location where I can see all the objects that various models can detect? Including the ipcam models?
Otherwise, how will I know which objects to specify for my chosen model?
 
Last edited:
Is there a central location where I can see all the objects that various models can detect? Including the ipcam models?
Otherwise, how will I know which objects to specify for my chosen model?

So for all Mike's customer models you can find the table on the first post here Mike's Custom Models

The yolo models have loads of objects (79), if you look here at the source code you can see them all listed yolov5 coco.yaml
 
Last edited:
  • Like
Reactions: SimonM and Corvus85
@Corvus85 - There are more yolov5 models than what is included with the CP.AI install, they can be used as custom models but unless you're rocking some serious hardware (3080/3090 etc) and also don't care about power consumstion I think there a bit overkill, which I way I assume the CP team them out. Aanyway they're here if you want to take a look yolov5 pretrained models
 
Is there a central location where I can see all the objects that various models can detect? Including the ipcam models?
Otherwise, how will I know which objects to specify for my chosen model?
You may want to start in page 1 of this thread and work your way forward. I know it’s a bit of reading but a lot of these questions are answered already and you will gain a bit more understanding of the entire project as well.
 
  • Like
Reactions: CrazyAsYou