Open leahcornelius opened 1 year ago
Hi, it was a quite old project. But at that time, running an object detector with TensorRT and processing one depth stream was around 8-10 fps. But we were using an operating system that is quite minimal, no GUIs, etc; which was heavily optimized.
Hi there, thanks for taking the time to out this together. I am currently looking at different boards to process my kinect (v2) data. I’m considering an intel NUC (i5 core based) some form of ODROID or a jetson nano (2 or 4gb). I was wondering what sort of performance you got when using a jetson nano as instructed here?
I will be using the data for SLAM/ mapping with a robot and I will need both RGB as wel as the depth streams. It doesn’t need to be at the 30fps level the kinect can max at, but needs to be decently high (15-20). I was originally planning on using a rpi 4b 2gb board but while it looks like it is quite simple to get running, the performance is limited to under 5fps which I doubt will work with my mapping setup.