This article was originally published on July 31, 2022.
Combining artificial intelligence with additional industry applications has always been a priority for acceleration solutions by Xilinx and its ISV ecosystem. Today, on the Xilinx App Store, you will be able to easily obtain an end-to-end commercially deployable solution for facial recognition, from Aupera’s , built on Xilinx Alveo™ accelerator cards and Zynq® SoCs, delivering, industry leading recognition accuracy and low latency with unsurpassed performance at a low hardware investment, assuring best-in-class TCO.
Aupera Smart Retail supports video feeds from multiple full HD cameras for retail object identification and classification purposes for loss prevention applications. The unique proposition of this solution is its ability to run multiple inferencing models concurrently and deliver accurate results at deterministic low latency, resulting in industry-leading low Total Cost of Ownership (TCO). It supports up to 32 channels of 1080p streams on a Boston Stream AI appliance populated with one Alveo U30 and one Alveo U50 accelerator card. The number of channels (video streams) and concurrent AI models can be increased by adding more Alveo accelerator cards (up to seven) to a single Stream AI appliance.
The application pipeline operates on top of Xilinx Video Machine Learning Streaming Server (VMSS) framework, delivering end-to-end pipeline acceleration management by running pre- and post-processing as well as inferencing on the FPGA fabric, delivering about 75% lower latency than a GPU-based system in a typical large retailer 32-camera store deployment.
The best part is that the solution is offered for free evaluation on Xilinx App Store. You only need to login to the Xilinx App Store Portal, obtain an account access key, and download the docker image for free evaluation.
More resources: