CNN-using-HLS

Directory structure:

This package contains the following directories:

• modules/ - directory used for the development and testing of individual HLS functions that are used for the CNN implementation
• nnet_stream/ - directory containing the C++ source and testbench
• py/ - directory containing the python code used to train the neural network

Generate Vivado HLS project:

Each directory contains gen_proj.tcl that can be used to setup te Vivado HLS environment.

To generate the project for the main CNN implemention follow the steps bellow:

1. Clone the repo: git clone https://github.com/amiq-consulting/CNN-using-HLS.git
2. Go to nnet_stream directory: CNN-using-HLS/nnet_stream/
3. Generate the project: vivado_hls -f gen_proj.tcl

Software used:

OS: Ubuntu 18.04.03 LTS

Vivado HLS 2018.3 - Simulation results and Synthesis

Python libraries

• numpy - version 1.15.1
• tensorflow - version 1.9.0
• sklearn - version 0.0
• scipy - version 1.1.0

Blog post

https://www.amiq.com/consulting/2018/12/14/how-to-implement-a-convolutional-neural-network-using-high-level-synthesis/

License

The application is available for free under the Apache License 2 We choosed this type of license to allow ANYONE use the materials as they please.