LexSet, a visual AI solution for object recognition and visual search, recently saw its efforts recognized by the AR/VR community with awards at the AWE USA and GS1Connect conferences.
The company won the inaugural Startup Pitch Competition at GS1Connect USA June, and the Startup to Watch Auggie Award at AWE USA in May. Both awards come with cash prizes and recognition from leaders within the AR and VR industries.
At AWE USA, LexSet was one of 18 finalists selected from more than 200 applicants. Nominees included everything from mixed reality creation tools, to games, to real-time training simulations.
At GS1Connect, finalists were evaluated on their product’s originality, usability and potential societal impact, as well as their overall product presentation.
LexSet CEO and co-founder Leslie Oliver Karpas said the awards would help the company expand its footprint and break into new sectors and industries.
“We are looking for additional partners in the robots and industrial space so we can do more part recognition and robotic navigation projects,” Karpas said. “There are Issues with the accuracy of computer vision. We want to make every computer vision system more accurate.”
The Future of Computer Vision
Karpas founded LexSet after working in architecture and 3D printing. He saw that machine vision could have uses far beyond architectural models. The company now provides a machine learning AI solution that can help train computer vision in any industry.
“Everything that has a camera on it will be running some sort of computer vision AI within the next five years,” Karpas said. “That creates a huge need for training data. Our long-term goal is to create a platform that would allow non-AI experts to create training data for machine learning.”
The company’s first success came in the contract furnishings industry, working to help brands streamline the competitive bidding process. They did that by “looking at what products of their competitors are on a bid, and helping them save 1000s of hours to instantly find their own equivalent products for their counter bid,” Karpas explained.
For example, if an architect wanted to see if a brand carried a piece that looked like something they liked in a given photo, they can log into the portal to search for it. This eliminates the need to manually look through product catalogs and websites, Karpas explained.
“In short, we’re good at recognizing and describing objects in images and real-time space,” he said. “If there’s no good data to train with, we create it.”
Moving Beyond Furniture
LexSet is also working on training data for robotic vacuum cleaners. Karpas said LexSet’s efforts helped achieve 98% accuracy between floor types. The next challenge is spatial recognition.
“We’re working on helping robots understand where objects are in relation to each other,” Karpas said. “People tend to stand in front of furniture so you’d want to vacuum more there. Or tell Alexa to have the robot vacuum under the couch.”
The company also worked with a mining company that makes very detailed drill bits that are nearly impossible for human eyes to tell apart. Machine vision, however, was able to learn the differences, even after the drill bits became worn.
According to Karpas, the compound annual growth rate for the machine vision industry is close to 50%. He expects that 5G will make machine vision on mobile devices much more accessible, opening new avenues for using it.
“Even an iPhone 10 doesn’t have a tremendous amount of power on the device. There’s too much going on for a phone to handle,” Karpas said. “5G will let you do the work in the cloud and stream it to your device quickly.”
