TypeError: stat: path should be string, bytes, os.PathLike or integer, not NoneType tfc_end2end_example

[sakshigoel361]

Hi,
I see similar error as discussed in the thread #392
I am running the example bnn-pynq/tfc_end2end_example inside the docker using 'bash ./run-docker.sh notebook'. I am using the latest version of dev branch. While using the function DeployToZYNQ, i get the error about static path. If I take care of this using os.stat(target_dir), then the error changes to that directory '/home/xilinx/finn_tfc_end2end_example' does not exist.
I am able to ssh to PYNQ board with bash ./run_docker.sh and cd to the directory /home/xilinx/finn_tfc_end2end_example , so it does exist. I have created this directory using mkdir before calling DeployToZYNQ
Error1

Error2

[rpitonak]

Hello,

I think the problem may be that on FINN dev there is no PYNQ driver generated by default during the Zynq build and these notebooks were not adapted to this change (I can do a PR once the answer is accepted and it resolves this issue). Try to add this before the step you are running.

from finn.core.modelwrapper import ModelWrapper
from finn.transformation.fpgadataflow.make_pynq_driver import MakePYNQDriver

model = ModelWrapper(build_dir + "/tfc_w1_a1_post_synthesis.onnx")
model = model.transform(MakePYNQDriver(platform="zynq-iodma"))
model.save(build_dir + "/tfc_w1_a1_pynq-driver.onnx")

and then execute

from finn.transformation.fpgadataflow.make_deployment import DeployToPYNQ

model = ModelWrapper(build_dir + "/tfc_w1_a1_pynq-driver.onnx")
model = model.transform(DeployToPYNQ(ip, port, username, password, target_dir))
model.save(build_dir + "/tfc_w1_a1_pynq_deploy.onnx")

Hope it helps.

[sakshigoel361]

@rpitonak Thank you for your help. With the addition you suggested, the error indeed goes away. I am using ZCU104 and on running the further steps of the tutorial I realized that the board does not have PYNQ image (There is no /pynq folder and I get the error ModuleNotFoundError: No module name 'pynq' when executing driver.py script on the board).
I think after adding a PYNQ image maybe additional lines of code are not needed. The board has ZCU104 image from previous VitisAI experiments and in order to add PYNQ image maybe I need to flash the SD card again with PYNQ image. Is this image sufficient for FINN dev?

[rpitonak]

Yes, the preferred target platform for FINN is a board with PYNQ image.

I am not using ZCU104, but it should be supported by FINN out-of-the-box since it is defined in pynq_part_map:
https://github.com/Xilinx/finn-base/blob/dev/src/finn/util/basic.py

For the PYNQ setup of your board see these links (just for reference, you probably already found them)
SD card image: http://www.pynq.io/board.html
Getting started guide: https://pynq.readthedocs.io/en/v2.6.1/getting_started/zcu104_setup.html#zcu104-board-setup

[sakshigoel361]

@rpitonak The example works now till the end and I can try to modify it for my project. Thanks for your help :)

[superpi15]

My PYNQ-Z2 works well on end to end example now.
Thank you again for the solution.