How to slow down the imagenet live camera detection prediction?

Autonomous Machines Jetson & Embedded Systems Jetson Nano
#1

Hello all,

I got a jetson nano and I am running “./imagenet-camera googlenet” to start a live camera detection. However, the predicted classes are being predicted so fast that I cannot even read the full name of the class. Is there anyway to slow that down a little bit? I mean like every 2 seconds output a prediction.

I tried to change the file “/jetson-inference/imagenet-camera/imagenet-camera.cpp” but the only thing came to my mind is “sleep()” function which holds the camera frames for 2 seconds and that kills the accuracy. Is there a better way?

Thank you in advance!

#2

Hi gharibimo, I think what you could do is keep an N-second running buffer of the predicted classifications, and you select the most frequent classification from the past N-seconds (where value of N is up to you, depending on how quickly you need changes to occur - anywhere from 0.5 seconds to 2 seconds maybe).

#3

Thank you so much @dusty_nv

#4