Object detection on TX2 speed up by not using detection every frame?


I am trying to speed up my yolo based detection on the tx2; is it possible to not use the detection every single frame but only every second, third…? And use a more “simply”/faster approach in the “in-between-frames” like background substraction, segmantic flow, optical flow, or something else, without losing to much on accuracy?

Thank you! Regards!


You can check our deepstream SDK:

You can set the inference interval in the config file.

Thank you for the quick answer!

Yes I am familiar with SDK; however, I would like to speed up my yolov2 on your tx2 (at the moment I am at 13fps) with a good MAP. The goal is to make it faster and therefore the question;

is it possible to increase the speed by using the object detector only every 2nd or 3rd frame and insert another method in the “in-between-frames”? this way I could massively increase the speed, maybe even double it if the other method is really fast.
= could this in-between-method be deepstream sdk??

example: 1frame yolo, 2 + 3 frame SDK, 4 frame yolo etc…??

Thank you very much, would be really helpful to get your opinion on it, or maybe something like that already exists!



You can set the interval parameter.

interval=1 means do inference every single frame.
interval=2 means do inference every two frames.

You can update the interval value in configure file directly.