Lenet-5 on PYNQ via HLS

• Software Platform: Vitis HLS 2020.2
• Hardware Platform: PYNQ-Z2 (xc7z020clg484-1)
• FP32 model accuracy: 97.65%; Int8 PTSQ model accuracy: 96.88%
• Software fps: 25; Hardware fps: 417; 17 times acceleration
• AXI Stream not used. Expected in the future.

Repository Tree

├── bd/
├── jupyter/
├── README.md
├── report.pdf
├── src_fp32/
└── src_int8/

bd folder contains the reference block design.

jupyter folder contains the bitstream files, the hardware description files, and the reference jupyter notebook.

src_fp32 and src_int8 contains HLS source codes for float32 and int8 models.