View on GitHub

PYNQ-2.7-MNIST

PYNQ-Based MNIST with Tensorflow Lite

PYNQ-2.7-MNIST

If this project is constructive, welcome to donate a drink to PayPal.

Compiled version is targeting “Custom Board” XC7Z010-CLG484

MNIST Network

MNIST Layers

Baseline of the above network

Test loss: 0.204

Test accuracy: 97.64

Quantized MNIST network

Test loss: 0.206

Test accuracy: 96.73

Vivado Resource usage

XC7Z010-CLG484

LUT 6415 17600 36.448864
LUTRAM 934 6000 15.566667
FF 6125 35200 17.400568
BRAM 27 60 45.0
DSP 47 80 58.749996
BUFG 1 32 3.125