People can mask USPS AI detections

This model worked fine in deepstack, I am trying to figure out why it is not working with code project.... If I scan the image manually in Codeproject it picks up the usps logo but if done through blue iris it bypasses it for some reason...
 
This model worked fine in deepstack, I am trying to figure out why it is not working with code project.... If I scan the image manually in Codeproject it picks up the usps logo but if done through blue iris it bypasses it for some reason...
I thought that maybe it had been renamed like some of mikeluds models but it appears unchanged from looking at this post in the codeproject thread, have you already looked at this thread? You could try a search since it's a rather long thread:

 
  • Like
Reactions: kaltertod
I created an expanded delivery vehicle model building off of the hard work of sstratoti and with help from mikelud1. The custom model is trained to recognize amazon, dhl, fedex, garbage, homedepot, ikea, lowes, target, uhaul, ups, usps, and walmart.

Sample captures: delivery truck images
Download link: delivery.pt (version 1.0 1.1 1.2 1.3 1.4)

Update: Uploaded a version 1.1 to the same link. I was getting some false positives with ups so I added more images and did additional training. Newer model is also smaller and faster.
Update: Uploaded version 1.2 1.3 with additional training on images for usps.
Edit: I'm also working on adding uhaul and perhaps garbage (truck), though it seems to be prone to false positives on garbage trucks.
Update: Uploaded version 1.4 that now has support for uhaul and garbage. For now, I would set detection to 90% on garbage to prevent false positives. If anyone would like to donate some images of garbage trucks, I'll add them and retrain the model.
Thank you for your effort on this, it is REALLY appreciated. Is it too much to ask that you place this on github.com so we can track updates?

I'm looking to implement this soon, again, thank you!
 
  • Like
Reactions: brianegge
I thought that maybe it had been renamed like some of mikeluds models but it appears unchanged from looking at this post in the codeproject thread, have you already looked at this thread? You could try a search since it's a rather long thread:

Searched the whole thread and the only references to USPS.pt in that thread are the ones that I made stating that I am unable to get it working..... might try a camera clone and see if I can get it working that way.
 
  • Like
Reactions: gwminor48
Interesting enough I did get a hit with it after deleting the camera and redoing all the settings on the camera. However it detected it from further away across the street. This is strange behavior, as before it would detect the logo on the same camera even with the door open from the close side of the street. The only change that has been made is going from Deepstack to Codeproject AI
 

Attachments

  • DrivewayE.20221004_111720542.10.jpg
    DrivewayE.20221004_111720542.10.jpg
    1.6 MB · Views: 37
Thank you for your effort on this, it is REALLY appreciated. Is it too much to ask that you place this on github.com so we can track updates?

I'm looking to implement this soon, again, thank you!
I have considered it, especially if I make more updates to delivery.pt. Anyone have a quick guide on doing that?
 
  • Like
Reactions: gwminor48
@VideoDad I am using delivery.pt v1.4 with BI 5.6.3.0 + CodeProject.AI 1.6.6.

Your custom model is excellent. I consistently get accurate detection for ups, usps, dhl, fedex and amazon---which are the labels I care about.

EXCEPT for the situation shown by image below. When the USPS guy leaves his door half open, I get nothing. I can try to whisper to him to close his door when he delivers mail to my home but he's going to think I'm weird, plus, we're not really friends.

Any chance you can include my image in your next training run so it can detect usps in this situation?DrivewayDelivery 2022-10-27 10.29.30.727 AM.jpgAI analysis.png
 
  • Like
Reactions: gwminor48
@VideoDad I am using delivery.pt v1.4 with BI 5.6.3.0 + CodeProject.AI 1.6.6.

Your custom model is excellent. I consistently get accurate detection for ups, usps, dhl, fedex and amazon---which are the labels I care about.

EXCEPT for the situation shown by image below. When the USPS guy leaves his door half open, I get nothing. I can try to whisper to him to close his door when he delivers mail to my home but he's going to think I'm weird, plus, we're not really friends.

Any chance you can include my image in your next training run so it can detect usps in this situation?View attachment 144006View attachment 144008
The model does include many images of open doors, but I can certainly add one more. I'm busy with Halloween preparation but I'll do some investigating early next month.
 
I'm having trouble with CodeProject.AI picking up the logos in my 2.88mm camera. When I use CodeProject.AI Explorer with screenshots, it won't find anything. If I crop the screenshot and try again, it finds the USPS and Amazon logos perfectly. See screenshots.

Any ideas? Do I need a camera with more zoom?
 

Attachments

  • Amazon-1.PNG
    Amazon-1.PNG
    365.2 KB · Views: 35
  • Amazon-1-cropped.PNG
    Amazon-1-cropped.PNG
    360.5 KB · Views: 33
  • USPS-1.PNG
    USPS-1.PNG
    375.6 KB · Views: 29
  • USPS-1-cropped.PNG
    USPS-1-cropped.PNG
    427.5 KB · Views: 27
I'm having trouble with CodeProject.AI picking up the logos in my 2.88mm camera. When I use CodeProject.AI Explorer with screenshots, it won't find anything. If I crop the screenshot and try again, it finds the USPS and Amazon logos perfectly. See screenshots.

Any ideas? Do I need a camera with more zoom?
Also what impacts detection is the model size selected, below is the code showing the resolution settings for each of the model sizes. This setting also adjusts the custom model resolution.

Python:
# see https://github.com/ultralytics/yolov5 for resolution data
    MODEL_SETTINGS = {
        "tiny":   Settings(STD_MODEL_NAME = "yolov5n", RESOLUTION = 256), # 640
        "small":  Settings(STD_MODEL_NAME = "yolov5s", RESOLUTION = 256),
        "medium": Settings(STD_MODEL_NAME = "yolov5m", RESOLUTION = 416),
        "large":  Settings(STD_MODEL_NAME = "yolov5l", RESOLUTION = 640),
        "huge":   Settings(STD_MODEL_NAME = "yolov5x", RESOLUTION = 640)  # Not yet included
    }
 
  • Like
Reactions: gwminor48
Also what impacts detection is the model size selected, below is the code showing the resolution settings for each of the model sizes. This setting also adjusts the custom model resolution.

Python:
# see https://github.com/ultralytics/yolov5 for resolution data
    MODEL_SETTINGS = {
        "tiny":   Settings(STD_MODEL_NAME = "yolov5n", RESOLUTION = 256), # 640
        "small":  Settings(STD_MODEL_NAME = "yolov5s", RESOLUTION = 256),
        "medium": Settings(STD_MODEL_NAME = "yolov5m", RESOLUTION = 416),
        "large":  Settings(STD_MODEL_NAME = "yolov5l", RESOLUTION = 640),
        "huge":   Settings(STD_MODEL_NAME = "yolov5x", RESOLUTION = 640)  # Not yet included
    }

So I need to adjust the setting here?
 

Attachments

  • BI-AI-settings.PNG
    BI-AI-settings.PNG
    112.7 KB · Views: 45
The model does include many images of open doors, but I can certainly add one more. I'm busy with Halloween preparation but I'll do some investigating early next month.

Hi VideoDad, I have the same issue, curiosity, what is the status of it? Thank you for the great job.
 
I know the values are a little high, I'm trying to figure why but this is what AI can see.

Thanks
 

Attachments

  • 2023-01-07 10_01_47-.png
    2023-01-07 10_01_47-.png
    649.9 KB · Views: 22
I know the values are a little high, I'm trying to figure why but this is what AI can see.

Thanks
If you would like, send me as many different photos as you can of times when it isn't working, especially open doors. I'll add them to my training model. I'm thinking the angle might be the other issue.
 
  • Like
Reactions: actran