Newbie question: Using the Jetson-Inference library, can I tell if there is something in the frame that cant be recognised?

I am using my Jetson Nano to identify objects passing in front of the camera. I’ve trained by own model. The background in always static.
I want to collect images of objects the device can’t recognise, so I can store them for retraining later.
So, I have three states;
Background - do nothing
Object detected - log the details
Something in the frame, that isnt the background, but cant detect it - take snapshot and store for retraining

Is this possible ? Any pointers would be greatly received.




This can be approximated manually.

For example, you know there are K objects in the scene.
So you can write a script to store the frames whose detected objects are less than K.


This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.