NanoOWL vs PeopleNet

I am undertaking a multi-company project to create a UAV prototype for surveillance and security of a restricted area and visual inspection of telecommunications infrastructures. My role in the project is to develop the AI component for intrusion detection and the visual analysis of infrastructures.

I decided to use your NVIDIA Jetson Orin Nano Developer Kit. I was studying/researching possible intrusion detection algorithms when I discovered that you have NanoOWL and PeopleNet. However, I wasn’t sure of the differences between them. Can anyone explain the differences between the two? How could I then train the algorithms? Do you also have software to adapt the code to my project/objectives?


Welcome to the NVIDIA Developer forums! Your topic will be best served in the Jetson category.

I will move this post over for visibility.


1 Like

My question is regarding the differences between NanoOWL and PeopleNet. Even so, should be moved to Jetson Category?

I don’t know of a better location for your topic. The Jetson folks will know for sure.

1 Like


PeopleNet is a pure detection model and can be trained with our TAO toolkit.

NanoOWL is a Vision Transformer (ViT) model which is optimized from an opensource model on huggingface.

In short, PeopleNet uses the predefined labels but NanoOWL’s output is runtime generated.


So for my case I should use the PeopleNet, right?

@tiagoveracruzcm ,

Thanks for reaching out.

PeopleNet is a model that is specialized in detecting people. If you need to detect people only, it is likely the better choice.

NanoOWL is an “Open-Vocabulary” object detection model, that detects objects based on generic prompts like (“a chair”, “a dog”, “a hat”, etc.). This may be a good starting point if you don’t have a model for your specialized task, or have multiple classes of interest.

Hope this helps.


Thank you for your help.

I will use PeopleNet for people detection.

I still have 2 more questions:

  • Does run in Jetson Orin Nano? For a flight of 30 minutes and sending the images for a outside user…
  • Could I use the NanoOwl for infrastructure inspection? Can detect flaws or it isn’t prepared for that?