A trained Convolutional Neural Network implemented on ZedBoard Zynq-7000 FPGA. Link to YouTube Video(s): https://www.youtube.com/watch?v=xoB--RFfy6I&feature=youtu.be
Project name: BeeBoard
Date: 30-Jul_2018
Version of uploaded archive: 1
University name: ISTANBUL TECHICAL UNIVERSITY
Supervisor name: Berna Ors Yalcin
Supervisor e-mail: [email protected]
Participant(s):
Ilayda Yaman https://www.linkedin.com/in/ilayda-yaman-9bba0ab1/
M. Tarik Tamyurek
Burak M. Gonultas https://www.linkedin.com/in/burak-mert-gonultas-94b045b1/
Email:
gonultasbu [at] gmail [dot] com
Board used: Digilent ZedBoard Zynq®-7000 ARM/FPGA SoC Development Board
Vivado Version: 2018.1
Brief description of project: A trained Convolutional Neural Network has been implemented on an FPGA evaluation board, ZedBoard Zynq-7000 FPGA, focused on fingerspelling recognition.
Description of archive (explain directory structure, documents and source files):
CNN folder includes Vivado files
MATLAB_Code folder includes files to verify the results obtained by the Vivado- Behavioral Synthesis
Instructions to build and test project
Step 1: Go to CNN folder for Vivado files of the project
Step 2: Run Behavioral Synthesis
Step 3: Obtain results for the hardware design
Step 4: Compare it with MATLAB results by running the "CNN.m" file inside the MATLAB_Code folder
FAQ:
Q: Which dataset has been used for training the CNN model?
We used LSA16 Argentinian Sign Language dataset: http://facundoq.github.io/datasets/lsa16/
Q: What do the output values correspond to?
See http://facundoq.github.io/datasets/lsa16/.
Q: Are the weights shared in this repo reliable?
Nope.
Q: Are the designs reliable and do you think it's a good idea to just copy and paste into my design?
I wouldn't bet on it.
Q: Should I expect solid results when I run the project?
Can't promise. Please understand that this was a class project done by 3 undergrad students. But still we worked hard on it and were satisfied with our designs at the end of the project. So feel free to use it as a reference.