Nearly all intelligent devices that perform computer vision, speech recognition, and signal processing tasks now use Neural Networks (NN). For the aforementioned applications, the efficiency and accuracy of Neural Networks have advanced to the point that researchers consider them to be more accurate than the conventional algorithmic approach. However, there are only a few hardware devices available that can be used to implement and deploy such Neural Network solutions at the edge for high-speed real-time analysis.

This product solution guide demonstrates the Binary Neural Network (BNN) and Quantized Neural Network (QNN) on Avnet's Ulta96-V2 using Xilinx PYNQ overlays. The users will implement image recognition applications such as Road Traffic Sign Detection and ImageNet Animal Identification using neural nets. This project explicates how to implement a hardware-based high performance acceleration model in an embedded processing AIoT edge application rather than a software implementation that has its own limitations.

What you will learn:

• Heterogeneous All Programmable Devices for Neural Networks
• Understand PYNQ and PYNQ overlays
• Explore the best avenues to get started with PYNQ and Ultra96-V2
• Neural Network and Architecture
• Challenges in Implementing Neural Networks
• Setup Ultra96-V2 hardware to boot and use PYNQ Framework
• Python Jupyter Notebook running on the Ultra96
• Design Example I - Road Traffic Sign Detection
• Design Example II – Animal