Purchasing furniture can be a pain in the neck. From measuring the various areas of one’s space to determine what furniture might or might not fit where, to comparing different colours and textures using tiny fabric swatches and so on. What’s more, even after the time-consuming shopping process is completed, many consumers end up disappointed with the furniture they’ve just bought once they actually see it in their room.
New York-based AI startup LexSet.ai wants to make furniture shopping as simple and fun as playing a kid’s video game, with an AI-powered plug-in interior designer that “analyzes your home, identifies the items in your space down to the SKU level, provides personalized product recommendations that are optimized for your space and reflect your taste.”
LexSet’s B2B solution is aimed at furniture and home supply retailers. With LexSet’s API integrated, a furniture shopping application can identify the objects in a smartphone camera viewfinder and search its directory for recommended furnishings based on elements such as names, tags, categories and styles. The application will then measure the spatial relationships in the scene, and use augmented reality to show how different products would appear in the user’s actual space.
Product matching and recommendations are not novel AI features — Alibaba and Google have already enabled generic image search allowing customers to find images or products containing similar features. LexSet however is “incredibly focused on home and offices,” as company Co-Founder and CEO Leslie Oliver Karpas told Synced. “We’re the only ones who are working very specifically to the furniture space, getting down to the SKU level.”
Karpas is a serial entrepreneur who founded the multi-million-dollar startup Metamason, which produces 3D-printed medical devices. Nine months ago, he started LexSet with CTO Francis Bitonti, a designer known for merging cutting edge digital design and manufacturing technologies and the founder of New York-based Studio Bitonti; and CBO Azam Khan, who was Director of New Ventures at Seattle-based research lab and incubator Intellectual Ventures.
“We spun out of Intellectual Ventures, and they gave us an exclusive license to a large set of patents all around context recognition and spatial search,” says Karpas. According to the Economist, Intellectual Ventures has filed more than 3,000 patents and acquired 70,000 more over the last 10 years, surpassing the numbers of Google, Toyota or Boeing.
Currently, LexSet is fully focused on its generic computer vision system, which detects furniture and reflects element results. The first step was accessing the databases of various big furniture brands to collect furniture data, particularly 3D CAD (computer aided design) files .
Theses files are then transformed into training data using procedural algorithms — similar to the technique widely used in film production — to generate colours and lifelike textures and map them onto an object; and CGI to simulate different backgrounds, lighting conditions and camera angles.
LexSet has so far packed almost 70,000 objects into its Convolutional Neural Network training dataset.
This generic computer vision system can also be applied to insurance, an industry that LexSet plans to target next. Says Karpas, “for those people who are going into the field after there has been a flood or a fire, we can simulate damaged objects in our database. So an adjuster can go to a disaster site and quickly identify the damage in that environment.
LexSet will also roll in placement recommendation by the end of this year. Developers will leverage urban planning and computer graphics to analyze point sets and extract space metrics, including adjacency, visual, and spatial hierarchy.
Karpas and LexSet are hoping to simplify furniture shopping by showing people what their new living room or a kitchen would look like in an automated and instant way. “When it comes to making furniture for space, even very smart people get very anxious. We are trying to be a player in AI interior design.”
Journalist: Tony Peng | Editor: Michael Sarazen