MikeLud1
IPCT Contributor
Hopefully today they will have the prerelease out for testing and if all testing goes well they will do the release.
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/
- Thread starter fenderman
- Start date
I am completely failing at adding the Custom Models from here - GitHub - MikeLud/CodeProject.AI-Custom-IPcam-Models
I have the Docker instance installed and it's working well, however, I'd like to ignore certain animals like cats but I can't figure out the path to use in Blue Iris.
I put the IPcam-combined.pt file into the C:\CustomModels\ folder on my Blue Iris machine, however, I then get this message repeated over and over in the CodeProject.AI log:
16:49:59: Object Detection (YOLOv5 6.2): /app/AnalysisLayer/ObjectDetectionYolo/custom-models/IPcam-combined.pt does not exist
16:49:59: Object Detection (YOLOv5 6.2): Unable to create YOLO detector for model IPcam-combined
Any help would be greatly appreciated.
The machine that the Docker image is running on does NOT have a GPU which is why I installed the CPU only version of CodeProject.AI when I started messing with it today.
I have the Docker instance installed and it's working well, however, I'd like to ignore certain animals like cats but I can't figure out the path to use in Blue Iris.
I put the IPcam-combined.pt file into the C:\CustomModels\ folder on my Blue Iris machine, however, I then get this message repeated over and over in the CodeProject.AI log:
16:49:59: Object Detection (YOLOv5 6.2): /app/AnalysisLayer/ObjectDetectionYolo/custom-models/IPcam-combined.pt does not exist
16:49:59: Object Detection (YOLOv5 6.2): Unable to create YOLO detector for model IPcam-combined
Any help would be greatly appreciated.
The machine that the Docker image is running on does NOT have a GPU which is why I installed the CPU only version of CodeProject.AI when I started messing with it today.
Hi,I finally decided to give codeproject AI a shot. but it does not seems to be detecting people.I am not seeing any events being logged on blueiris, even though when people walk in front of camera object detection rectangles appear over them. Codeproject AI explorer logs indicate it is working as expected. What am i doing wrong?
4:20:02 AM: Latest version available is 1.6.7-Beta
4:20:05 AM: Latest version available is 1.6.7-Beta
4:20:12 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'detect' took 204ms
4:20:12 AM: Object Detection (YOLO): Detecting using ipcam-combined
4:20:12 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'custom' took 117ms
4:21:06 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'detect' took 178ms
4:21:06 AM: Object Detection (YOLO): Detecting using ipcam-combined
4:21:06 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'custom' took 98ms
4:22:13 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'detect' took 224ms
4:22:13 AM: Object Detection (YOLO): Detecting using ipcam-combined
4:22:13 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'custom' took 117ms
4:23:06 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'detect' took 229ms
4:23:06 AM: Object Detection (YOLO): Detecting using ipcam-combined
4:23:06 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'custom' took 113ms
4:20:02 AM: Latest version available is 1.6.7-Beta
4:20:05 AM: Latest version available is 1.6.7-Beta
4:20:12 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'detect' took 204ms
4:20:12 AM: Object Detection (YOLO): Detecting using ipcam-combined
4:20:12 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'custom' took 117ms
4:21:06 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'detect' took 178ms
4:21:06 AM: Object Detection (YOLO): Detecting using ipcam-combined
4:21:06 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'custom' took 98ms
4:22:13 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'detect' took 224ms
4:22:13 AM: Object Detection (YOLO): Detecting using ipcam-combined
4:22:13 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'custom' took 117ms
4:23:06 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'detect' took 229ms
4:23:06 AM: Object Detection (YOLO): Detecting using ipcam-combined
4:23:06 AM: Object Detection (YOLO): Queue and Processing Object Detection (YOLO) command 'custom' took 113ms
Looks like you have AI working. Show us a screen shot of the BI camera's AI settings.
in the camera AI settings, in the custom models field, just use ipcam-combined You also might have a zone issue. Crossing zones can be flakey. I just use normal motion detection and let the AI work.
Snowmageddon is coming end of the week- hopefully 2.X is released by then since everyone in the northern US will be snowed in and can give us something to do LOL
Last edited:
Right. I've got local PD asking for my setup right now too due to mail, package, and vehicle item theft. I'm waiting on this release to test before sending them the info. This would definitely keep me busy during all the snow
It seems like working ,but detection is not accurate. I am using it in india and possibly detection not working due to different attaire worn by people. Is there anyway to train based on custom images.Or does the accuracy improve over time?
MikeLud1
IPCT Contributor
You can train your own model using custom image, see the below links on how to do this with YOLOv5.
Train Custom Data · ultralytics/yolov5 Wiki
Tips for Best Training Results
YOLOv5 in PyTorch > ONNX > CoreML > TFLite. Contribute to ultralytics/yolov5 development by creating an account on GitHub.
github.com
MikeLud1
IPCT Contributor
At a minimum 1,000 images the more images you have the better the model will be
you have to download it manually and put it in the folder with the other models.....then you can use it
CrazyAsYou
Getting comfortable
@MikeLud1, putting the time and effort this would require to one side, do you think there would be any advantage in people training their own custom models based on images purely taken from our own CCTV recordings, of course I’m thinking that you’d need to use all your cameras and also have a good selection of everything you’d want to identify from each one at day and night but it’s something I’ve been thinking about for a while now. My logic is that it would include how people and animals actually look for our setups, camera angles, lighting etc
One will absolutely have better results using their own images from their own CCTV.
I don't think someone has to do that as their first course of action, but if the available models are not successful or the accuracy is not there for a particular user or someone wants to identify something not available in a model, then training your own model with your own images makes the most sense, especially once you get beyond human or vehicle objects.
I think the biggest mistake people make is trying to do too much with one field of view, so if a model has say a squirrel in it and all the trained images are up close images but someone is trying to use it on a camera that is 20 feet high, the object isn't going to look the same and the detection and accuracy will suffer.
Last edited:
One thing I've wondered about is if BI can draw a box around the area in which it sees motion, why can't it send just what is in that box to the AI for analysis? I have a camera that looks down our driveway to a cul-de-sac.
BI often triggers motion on vehicles in the cul de sac, but when CPAI or DS processes the entire image rarely is a vehicle identified. I just took a still from a cancelled motion capture (nothing found) and cropped it down to just the cul-de-sac and had CPAI explorer scan that cropped image it came back with a 93% match as a vehicle.
Last edited: