I’m starting out with Hailo8 and trying to get a very simple inference pipeline going using gstreamer on Windows. Tappas looks like it’s the way to do this - however it’s only available on linux.
Eventually we’ll be using c++, but I’d like to see if I can whip something up as a quick demonstration.
I’ve seen other tickets describing something like:
I recommend using the Hailo AI Software Suite Docker under Ubuntu for quick demonstrations. You only need to install the PCIe driver outside of the Docker and have everything available including an example detection pipeline.
That element does not exist — it’s likely a typo. In GStreamer, queue elements are used to decouple different parts of a pipeline into separate threads with buffer support.
Please check the Tappas User guide for all GStreamer elements provided by Hailo except the HailoNet. For HailoNet please read the HailoRT User Guide. It includes instructions for the hailonet element under Windows.
You’re right – docker sounds like a good way to go to get something up quickly – I’d been avoiding it as I wasn’t thinking I’d get to far showing anything visually.
One of the initial challenges I faced was that I didn’t copy the hailonet plugin dll/libs into the gstreamer directory correctly first time and so gstreamer had black-listed it.
Once I’d deleted the gstreamer cache it discovered it properly and away I could go again.
So I can now pick up a file or camera and feed a video stream into the hailonet plugin using a yolo hef file from the zoo.
Given that our target platform is windows, it would be nice if I could also use the hailofilter and hailooverlay elements also.
Whilst unsupported, do you know if there has been any work done on compiling these for windows that I could leverage?
I’m not aware of any existing efforts to compile these for Windows. If you decide to give it a try, we would be interested to hear how it goes. Your experience could be helpful to others in the Community.