search

Xilinx / BNN-PYNQ

Quantized Neural Networks (QNNs) on PYNQ
https://xilinx.github.io/finn/
BSD 3-Clause "New" or "Revised" License
671 stars 306 forks source link

After bitsteam generation #9

Closed mgepner closed 7 years ago

mgepner commented 7 years ago

Once the bitstream is generated, do you program the FPGA with the bitstream using Vivado? How do you have the FPGA take an MNIST file and measure its performance? I have the bitstream created, as well as Linux running on the PYNQ and the mnist_parameters.npz/binparam-lfc-pynq created. What are the next steps?

giuliogamba commented 7 years ago

In order to use your generated bitstream, you can follow step 6 in HW design rebuild available at https://github.com/Xilinx/BNN-PYNQ/blob/master/README.md.

The binary parameters should be moved in pip_installation_path/bnn/params/

mgepner commented 7 years ago

Okay, I programmed the PYNQ now using the bit file found in bnn/bitstreams. The PYNQ is also running Linux. Are you saying the binary parameters in /bnn/params/ need to be moved onto the PYNQ?

giuliogamba commented 7 years ago

Yes, the parameters are stored on the PYNQ board after pip installation, and if a new dataset is trained those should be moved in the same path (_pip_installationpath/bnn/params/)

mgepner commented 7 years ago

The PYNQ is programmed with a bitstream through Vivado and has the necessary thres/weights files in /bnn/params/mnist now. How do you run the BNN on the PYNQ?

giuliogamba commented 7 years ago

No, the PYNQ in not programmed through Vivado. Please refer to the "Using Overlays" paragraph in http://pynq.readthedocs.io/en/latest/16_creating_overlays.html. To see how the BNN are run, please check the bnn.py class in the bnn subfolder