Ver 6.0.9.6 Built-In AI - using Delivery or USPS.onnx custom models failing to Detect usps

E

ELA

n3wb
Dec 11, 2021
20
5
home
Would like to start by thanking this community for all the guidance it has provided over the years, saving me many hours of experimentation.MailCam AI Cancelled Alert2.jpgMailCam AI settings2.jpgV6 camera Global AI settings.jpg

I recently upgraded from BI Ver 5.5.9.6 using Deepstack AI, to Ver 6.0.9.0 and using the built-In AI. The transition went amazingly easy at first.
Over the course of a week I found several things I needed to correct, mostly for new configurations due to the large leap I took in versions.

I am now struggling and unable to get the Delivery or USPS.onnx custom models to work properly. In the AI inspector I can see them being checked
but never any detections of "usps" at any percent level of confidence.

Have many here been successful with these custom models in these newest versions? The primary models are working fine, just neither of these custom models.

I have a clone running only these custom models that worked very well in BI V5 and Deepstack previous to this upgrade.

I downloaded Delivery.onnx with the installer package, and USPS.onnx from this site. The USPS.onnx file appears to be from yr2024.
Tried them separately and together.

I am seeing an AI:error 1460 in the log from time to time when the failing clone is active detecting motion. I tried every suggestion
I read about for that error, HA- is presently off.

I can see the computer (Intel(R) Core(TM) i7-10700 CPU) is now more taxed than with V5 BI but would hope still sufficient.

I have read the help files over and over as well as posts here. Did a lot of experimentation with changes without success.
I will attach several screen shot of my configuration. If anyone can help me with what I have missed that would be much appreciated.

I have not tried going backwards to the most stable current ver. 6.0.8.6.
 
I tried the delivery model about 1 month ago w/ no luck. I went ahead and just loaded it again. I am now on 6.0.86.

I'll post back with how well it works. I should have fedex, amazon deliveries today or tomorrow.
 
Just had Amazon deliver, but they delivered in a regular car so delivery model didn't work. AI picked up the car and person though :)
 
I just had a fedex and 2 UPS trucks go by and an amazon truck pull in my driveway and the delivery model didn't fire.
 
Last edited:
mvoss said:
I just had a fedex and 2 UPS trucks go by and an amazon truck pull in my driveway and the delivery model didn't fire.
Click to expand...
Thank you for your feedback and for saving me from needing to try regressing versions.

My apologies for dual postings. This first one did not appear the next day after posting. Then when I posted again today they both got published for some reason.

I hesitated upgrading from V5 and deepstack to avoid needing to goto CodeProject AI. I did not want to loose my usps detection.
I wonder if it is worth installing CodeProject with the new Version 6 to get usps working again?
 
Understood and thanks. I did try updating to 6.0.9.8 but no joy.
I tried the IPcam-dark model just to see if it detected as an "additional model" and that worked ok.

Wonder if a newer version of Delivery.onnx, tested with this built-in AI, may be available at some point, or if a BI version update is required?
 
Last edited:
I was testing the same delivery model as Tinman has shown and had also tried the USPS.onnx dated 4/28/24 (downloaded directly an old post on this site).
I do not know how to show the metadata, can you please share how to find that information for models?
 
I've been using it with fairly good results, I did find I had to capitalize USPS to make it work for me. I'm on 6.0.8.6
 
I'm still on BI 5.9.9.64 with CPAI using the YOLOv8 model (delivery.pt) so can't easily test the ML/NET/ONNX version in BI6. The YOLOv8 NET model (delivery.onnx) is built directly from the YOLOv8 model (delivery.pt), so my assumption is it should work unless BI 6 handles custom models differently than CPAI.

For determining a cause, can anyone provide some details on whether you had it working with CPAI in BI 5, then upgraded to BI 6 (still using CPAI) and was the delivery model still working? Is it only when switching the built-in AI processing that it stops working? What if you switch it back to CPAI using YOLOv8?

What else can you tell me about your set-up? Can you show me the camera setting screen? Did you add the 6 object classes to the confirm field? Are they also added to the objects that get listed as Vehicles?

I did also build a YOLO11 NET version of delivery.onnx (AI-Models/Delivery/YOLO11 NET at main · VideoDad/AI-Models), but I'm not sure if the built-in AI supports YOLO 11 or not. Anyone know? Or would someone want to try out this one in case? Also, any exact error messages could be helpful.
 
I normally don't use the delivery models, but I can confirm BI6 by default can use Yolov8 and Yolov10. Ken stated that others may work fine but are not supported yet. I'm running Yolov26s on my system and it is very fast and works great. So, are we assuming you MUST run yolov8 in order to use the delivery model? So far, its looking like a no go using yolo26.

Screenshot 2026-07-24 190418.jpg
 
ELA said:
Would like to start by thanking this community for all the guidance it has provided over the years, saving me many hours of experimentation.View attachment 246626View attachment 246627View attachment 246628

I recently upgraded from BI Ver 5.5.9.6 using Deepstack AI, to Ver 6.0.9.0 and using the built-In AI. The transition went amazingly easy at first.
Over the course of a week I found several things I needed to correct, mostly for new configurations due to the large leap I took in versions.

I am now struggling and unable to get the Delivery or USPS.onnx custom models to work properly. In the AI inspector I can see them being checked
but never any detections of "usps" at any percent level of confidence.

Have many here been successful with these custom models in these newest versions? The primary models are working fine, just neither of these custom models.

I have a clone running only these custom models that worked very well in BI V5 and Deepstack previous to this upgrade.

I downloaded Delivery.onnx with the installer package, and USPS.onnx from this site. The USPS.onnx file appears to be from yr2024.
Tried them separately and together.

I am seeing an AI:error 1460 in the log from time to time when the failing clone is active detecting motion. I tried every suggestion
I read about for that error, HA- is presently off.

I can see the computer (Intel(R) Core(TM) i7-10700 CPU) is now more taxed than with V5 BI but would hope still sufficient.

I have read the help files over and over as well as posts here. Did a lot of experimentation with changes without success.
I will attach several screen shot of my configuration. If anyone can help me with what I have missed that would be much appreciated.

I have not tried going backwards to the most stable current ver. 6.0.8.6.
Click to expand...
Try cloning this camera and make an area of interest, being the image resolution is at 1520x2688 the image gets resized to 640x640 and the USPS truck is to small to identify.

1784942023017.png
 
Hello Mike and thank you for responding. I was not familiar with the AOI setting and I will follow your advice and try that.

Don't know if something similar to AOI was present in BI 5.5.9.6?
I had the very same clone/area working fine with the same resolution in BI5 and deepstack without any knowledge of AOI.

Use of AOI makes total sense so thank you for pointing me in that direction.
I will post results tomorrow after testing.
 
  • Like
Reactions: MikeLud1
I haven't checked lately as been too busy, but I recently went to this version and my AI wasn't working at all, 0 detections.
 
I was able to capture 2 instances with the new AOI implemented. I really like the way the AOI limits the viewing area making the objects of interest appear to be larger. Thank you for pointing that out to me.
Unfortunately I am not sure it helped with AI detection much? I did get one false fire of a truck taillight as usps. Not sure why it then failed to present any detections for the actual usps truck that "appears" rather clear in view. ( both attached).
Since the AI model resizes to 640x640 I wonder if I should enlarge my AOI to be at least 640 high rather than the 320 it is at now. While I am not interested in the additional area might it possibly help the AI?

I recently added a new camera that also has a view of the road. I am going to add a clone with AOI to that camera to see if it performs any better.
Thanks for all the input.MailCam usps fail.jpgMailCam usps fail2.jpg
 
You must log in or register to reply here.
Top Bottom