This is a project for accelerating MNIST classification using FPGA pynq board. We implemented streaming architecture using 1-bit quantization. As a result, our hardware is currently about 10x faster than using numpy for MNIST classification.
src: vivado HLS source code for main hardware
MNIST-accelerator-comparison.ipynb: An example for comparing the speed of hardware and numpy
preprocess.ipynb: with this file, you can preprocess weights or preprocess dataset for specific batch size
