Which type net does xavier suit best?

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
#1

I try to train a 2D detection , and run it in xavier.

For the reason different model performs not the same in different hardware.

which type model has the best performance?

the residual type like resnet or parallal type like inception?

Hope for an advise or tutorial page to introduce .

best wishes
Cheng

#3

Hi,

For Jetson, since the GPU source is limited, the parallel type may not have gain as it does on desktop GPU.
However, we don’t have a concrete answer for this since it depends on the real model architecture.

You can find some performance result for detector here:

Thanks.

#5