This repo contains the code samples for the "From Scratch: Neural Network inference on FPGAs" posts.
The examples have been tested on AWS F1 instances using the FPGA AMI for development. See the first post for detailed instructions how to run them.
The weights in the weights/
folder were generated running
$ python train.py --outdir weights --epochs 10
with the final output
Accuracy: 0.9370993589743589
Model predictions: [0 1 2 2 4 5 6 7 8 9]
Note that the last line outputs predictions on images with labels 0
, ..., to 9
.