Open GOBish opened 4 years ago
Hi GOBish. Have you found a solution to this problem? I'm recently looking at Tracking on Xavier NX, but it seems that other deep learning based methods require high computing power, many of them require 2080Ti/2080Ti+ GPU cards.
Hi @HigashikataZhangsuke , I have not been able to increase the fps of SiamMask on the XavierNX much, but have been using non-mask Siam object tracking on the XavierNX at about 15-17 fps.
Hi @HigashikataZhangsuke , I have not been able to increase the fps of SiamMask on the XavierNX much, but have been using non-mask Siam object tracking on the XavierNX at about 15-17 fps.
Ah, I see.Really thank you for your reply, I'd like to try this method.
@GOBish which tracking model did you find suitable for XavierNX? How the results compare to SiamMask model?
Hi @shyld , I never got the SiamMask model to work. I wound up using a standard (non mask) Siam tracker.
Hi @shyld , I never got the SiamMask model to work. I wound up using a standard (non mask) Siam tracker.
@GOBish Thanks for your reply. Could you please point me to the GitHub repo that you used for the standard Siam tracker?
@GOBish Which model you used from this repo?
I tried the siamrpn_alex_dwxcorr and siamrpn_mobilev2_l234_dwxcorr but still getting 3-4 fps so if you could guide that would be great help.
Hi,
I am running the demo.py on a Jetson Xavier NX and only getting ~4FPS.
I have verified it is running on the GPU.
I expected to get much higher frame rates - is this attainable on the device? Would I have to convert the model from pytorch to tensorrt? If so, do you know of any examples showing this implementation?
Thanks!