PYNQ DPU + Vitis AI examples notebooks for object Detection

Hi,

I have been trying to execute the DPU related examples provided in the Vitis AI Zoo on the PYNQ platform. The example notebooks that are available are generally for object classification. With a bit of searching, I came across the information in the searches that a few of the notebooks (related to object detection through YOLO and SSD) have been removed from the PYNQ DPU Github repo. Can anyone please help and direct me as to how I can get the Models in the Vitis AI zoo executed on the PYNQ ZCU104 platform? Any information in this direction will be very helpful.

Thanks in advance.

1 Like

I think there are instructions here? Have you tried? You will need a x86 machine to rebuild them yourself.

I have built the xmodel files necessary for the examples, but needed help to code up (jupyter notebook example) for providing data and obtaining back data from the PYNQ inference models.

Hi there,

The previous notebooks (YOLO) were using the DNNDK api, which has now been deprecated and replaced with the Vitis AI Runtime (VART). This is probably more of a Vitis AI question than a PYNQ one. Your best bet is to find a similar python application in the Vitis AI github repository (something like this). Hope that helps.

Thanks
Shawn

Will take this lead and give it a try. Thanks.

Hi, were you able to run YOLO on PYNQ without DNNDK?

I haven’t worked on yolo for now, but i was able to works with custom models of my own.

Are there any direct way of running VART from PYNQ without the pybind11? That is the only example I have found to access more than one layer output.

Never mind, I found it.
If someone comes across the same question when you do
import pynq_dpu
it imports vart and xil for you. Looking at dpu.py in the DPU-PYNQ repo the path has to be appended first to sys.path

1 Like