Nvidia announced new ways to put AI to work for losses through theft in grocery and other retail operations prior to the opening of the National Retail Federation Conference Jan 15-17.
Separately at NRF, Sony Semiconductor Solutions said it will demonstrate AI sensing through its Aitrios platform for making predictions about on-shelf availability of products and store optimization. It will launch Aitrios through Microsoft Azure Marketplace to US and European partners and customers. Aitrios-compatible AI cameras equipped with intelligent IMX500 vision are now available, Sony added.
Qualcomm is also showing a dozen applications at the NRF event to help retailers, including drone last mile delivery with TDK sensors.
Developers of retail applications can use new Nvidia Retail AI Workflows to act as building blocks in their designs. The workflows come with thousands of pretrained images of most-stolen products along with software to be plugged into existing store applications at point of sale checkout machines. With image scanning compared to a database of products, retailers can tell if a product being checked out by a shopper is getting charged the accurate price since some unethical shoppers have been known to substitute a low-priced item for a higher one.
Also, images in a database can identify products seen in multiple cameras used especially in stores in Asia and Europe where in the frictionless shopping experience is catching on. Frictionless shopping is when a shopper moves through a grocery and picks up an item such as meat or milk then stores the product in a basket, an action that may be recorded through a high-definition camera.
“There’s huge traction in frictionless shipping in Asia and Europe,” said Azita Martin, vice president of retail for Nvidia, in a briefing with reporters.
One way Nvidia’s workflow operates is by offering developers AI that recognizes hundreds of frequently stolen object like alcohol, but also in various sizes and shapes. With Nvidia Omniverse and synthetic data generation, independent software vendors can further train the AI to hundreds of thousands more products.
The workflow is further enhanced with a few-shot learning technique developed by Nvidia to identify and capture images of new products not in the database as products are scanned at checkout.
The workflow can be used with computer vision to provide store analytics on traffic trends and where the busiest parts of a store are located. These workflows are built on Nvidia Metropolis Microservices.
The workflows will function in servers from companies such as Dell or Lenovo that run Nvidia GPUs in the back of large stores, Martin said. Infosys said it will be using Nvidia’s new workflows comprised of pre-trained models for retail SKU recognition.
Martin said Nvidia-based algorithms will have 98% accuracy and help recoup massive losses through theft. The global retail industry estimates it loses $100 billion a year in loss, damage or theft of products, with about 65% due to theft. NRF has found theft on the rise with rising food prices.