PYNQ: PYTHON PRODUCTIVITY

Changes to be made in overlay to modify the BNN

Hi, I want to implement some research paper on PYNQ i.e. need to modify the bnn/qnn architecture (say the activation function, layers etc.). So, what changes should be made in the vivado block design to generate bitstream? and what all things will remain same (e.g. AXI)

Also, how can I get the block diagram of bnn in vivado project to get bitstreams? (tried with .tcl file from http://192.168.2.99:9090/tree/BNN-PYNQ-master/bnn/bitstreams/pynqZ1-Z2/ but it had version issues)

1 Like

I think you are asking about FINN - not PYNQ.

There is a FINN support community here:
https://xilinx.github.io/finn/community

Cathal

1 Like

Thanks for the reply.
But, can you help with modifying the hardware files according to applications?
As, right now, we have many pre-built models for different applications but no guidance for modifying those models (especially the vivado part) e.g. We have trained BNN for image mnist dataset but I need to classify some other datasets and hence need to train the model for it. Don’t know how to do it.

The same goes for other applications like signal processing etc.

1 Like

This forum is intended for questions related to PYNQ.
If you have general quesitons on hardware design it would be better to post these to the Xilinx forums.
If you have questions related to how to modify hardware generated by FINN, or applications related to FINN designs I think you would be better posting to the FINN support channel.

Cathal

1 Like