Accelerating CNN in hardware
Along with increasing of AI computing, software can't fulfill our expectation anymore ( in speed performance ). On the otherhand, using hardware to implement specific algorithm is getting popular. This project is based on this aspect, we tried to accelerate CNN computing through hardware.
Following is how our project distinguish hand-written digit
- Input one hand-written picture
- Move bitmap into hardware which was synthesised in FPGA ( PYNQ ) already. ( We use
pythonand PYNQ built-in function to achieve it) - Do convolutional computing in hardware
- Move output after convolutional computing through PYNQ API
- Do fully connective computing
- Finally, we get the result of recognization
- vivado
- verilog
- jupyter notebook
performance improvement
