Unofficial Guide to the Jetson TK1

[SamCarlberg]

Originally Posted by AMendenhall

Very informative. Thanks for the post.

Why not use the ethernet port on the Jetson to put data directly into NetworkTables where it can be accessed by both the RoboRIO and the DS? Also, why not use Java; that's what is recommended by WPIlib?

The Java bindings don't support CUDA. You have to use C++ to get GPU acceleration, which is the entire point of the Jetson.

Originally Posted by oshamaa

I think, not quite sure, that using if you want the fasted vision processing you would want to use OpenCV, and OpenCV does not have a native API for Java, Java would just slow it down. I would not use NetworkTables, because that would also slow the process down, and risks the max-bandwidth being met in matches. With this method we had the max frames per second last year, specifically it only used 25% of our Cores.

OpenCV does have an official Java API. It just doesn't include the CUDA functions. NetworkTables is asynchronous; it wouldn't slow down the vision processing. There'd be a bit of latency (up to 10ms-100ms depending on how you configure it), but uses very little bandwidth, so your final statement there is just plain wrong.