Pednet tolerance threshold not working

Hi guys, I love using jetson inference for my projects and I found ped-100 and multiped-500 to be very effective at detecting persons at a distance. However, they detect trees, chairs, etc as a person, and does not matter how high I set the threshold .5 .8 .99 they keep misinterpreting the shapes. This does not happen with mobile net or others. What can I do?

Hi,

You can try the other models that trained on a larger database for a better accuracy:

Thanks.

I see, thanks. Is it possible to run 2 objecet detection models or one object detection and one image recognition model in the same python script?

@cespedesk yes you can do that, however when you run net.Detect() you want to use the overlay='none' parameter because otherwise it will draw the bounding boxes on the image which could impact the processing of the other models on the same image. Or you can make a copy of the image first.

Also, for multiple model pipelines, you may want to check out DeepStream SDK for optimized performance.

1 Like

Thanks, very helpful.