FPGA AI Suite: Design Examples User Guide

ID 848957
Date 4/30/2025
Public
Document Table of Contents

22.3.5.1. [SOC] Layout Transform Considerations

Pixels are typically 8-bit integer values, and the FPGA AI Suite requires FP16 values. As well as the c_vector padding, the layout transformation module converts the integer values to floating-point values.

The S2M example is a video-oriented demonstration. For networks such as ResNet50, the input pixel data must further be manipulated with a "mean" and "variance" value. The layout transformation module performs basic operation of Y=A*B+C operation on each pixel to meet the needs of a ResNet50 graph trained for ImageNet.