Solving the computer vision
training data problem
by simulating the world in 3D.
At LexSet did not start out as a synthetic data generation company. Our original mission was to create a spatially-aware AI for interior design via mixed reality. A computer vision system that could recognize all of the objects in a room and make recommendations on what products would go well in the space, as well as how to rearrange what was there.
As we were exploring how to accomplish this goal, we amassed a gargantuan data set of furniture 3D assets; and discovered the power of synthetic data for powering visual search, object recognition, and spatial navigation. It then dawned on us that the power of synthetic data was far more vast than the initial problem we went out to solve- and like so many other startups, pivoted to spread the value of the tool we'd built to solve our own problems.
At LexSet we've created a better way, generating data synthetically from 3D environments to produce limitless quantities of
ideal image data sets, with perfect annotations,
free from bias. Yielding faster, more accurate AI
for less money.
The dirty secret in AI, is that those AIs require vast quantities of data annotated by humans; and AI is only as smart as the data it’s trained on.
Better AI Training
Training AI in the past:
waiting for photos to be taken and annotated
Slow & Painful
Training AI with LexSet:
Fast & Ideal
"Tomorrow's Tech Today: Takeaways from Augmented World Expo"
Charlie Fink, Forbes
"10 Fastest Growing Companies in Retail '19"
The Silicon Review
"LexSet: Spatial Understanding Powered by Synthetic Data"
Alexander Culafi, Venture Fizz
Astasia Myers, Redpoint Ventures
"AI company LexSet debuts visual search software"
Anne Flynn Ware,
"Where Artificial Intelligence Meets Interior Design"
"Home decorating gets a new AR toolkit thanks to Intellectual Ventures' new incubator"
Jonathan Shieber, TechCrunch