Home Dice How do you train the AI-powered checkout to recognize a product? In Vegas (and elsewhere) you roll it like dice

How do you train the AI-powered checkout to recognize a product? In Vegas (and elsewhere) you roll it like dice

0

When it comes to training in machine vision and AI-powered retail checkout systems, packaged goods and locally created food products are treated very differently.

At least, according to Mashgin, a manufacturer of contactless payment systems. Company spokesperson Toby Awalt said it was because another store in the network had likely already added this bag of crisps or goodies to its database of more than 10,000 items.

This is not the case when it comes to locally produced food products.

“CPG items, we have to do less and less because there is enough overlap” said Awalt, who introduced us to the system at CES 2022. “But for the dishes, we’ll do it every time.”

According to Awalt, adding a new menu item for a restaurant doesn’t take that since most cafeterias or restaurants only serve between 15 and 50 items.

“You can do it pretty quickly,” he said.

However, a new packaged product must be entered into the system from time to time. Each time this happens, the operator must position the package in several different positions to give the system enough information to recognize the product each time it appears under the camera.

Toby Awalt from Mashgin rolls the Häagen-Dazs

“I actually make dice rolls with the product,” Awalt said, tossing a Häagen-Dazs ice cream bar on the board.

According to Mashgin, the company recommends that the system capture 20 to 50 total positions of a product so that it can recognize the product from different angles and also distinguish between different variations within the same product line (such as two different flavors. ice cream or chips).

You can watch an overview of the Mashgin system below.