LexSet, better data for better vision. For computer vision AI used for object recognition and spatial navigation, the biggest problem is the lack of well-annotated training data. LexSet solves this by simulating the world in 3D, creating limitless, unbiased data; dramatically improving AI accuracy and ease of development.
We did $275k in revenue in 2019, and are on track for $350k+ in 2020. We have 6 customers and average revenue presently of $20k+ per month.
We're focused on interior applications of computer vision including service robotics, mixed reality, industrial part recognition.
Problem or Opportunity
Everyone knows that AI is a major growth industry, what people don't talk about is that AI is only as good as the training data it's fed. This is a particular problem for Computer Vision (TAM: $25B market) where sourcing and prepping adequate image data is time-consuming and expensive. Forbes has shown that AI Vision engineers waste 82% of their time on data prep, and The Economist is calling annotation a $2B-$5B market (SAM).
Solution (product or service)
nstead of sending images to click farms in south-east Asia, LexSet's approach is cheaper and faster: Training Data as a Service or TDaaS, uses 3D content to automate the creation of pre-annotated synthetic image data to train Vision AI models. The approach allows users to generate limitless amounts of training data on-demand; customizing the camera type, lighting conditions, occlusions, and materials in a training set purpose generated for each application. TDaaS has been specifically proven in making better AI vision systems for object recognition and robotic navigation.
Ai.Reverie and CVEDIA
Advantages or differentiators
We've all been focusing on different markets, with AI.Reverie focusing in on Defense, Autonomous Vehicles and Cashierless checkout and CVEDIA also focusing much of the same.
We've chosen a narrower focus on interior space use cases.
We sell training data subscriptions we call TDaaS. The subscription price scales with the length and size of the data commitment. Typically our contracts range between $30k and $150k.
Combination of SaaS and service work (for those teams that need us to be their outsourced AI solution).
LexSet has been working with a number of robot vacuum cleaner makers, a large German tool maker, a large US mixed reality studio, a major US contract furniture maker, and have seen great success in improving the quality of their Computer Vision AIs.
Money will be spent on
Moving our product from an enterprise service to a standalone dashboard for mass developer use for the creation of AI training data.
Offer for investor
This fall we plan to raise a $3M-4M priced Series A on a $12M-15M pre-money valuation.
That the computer vision market doesn't grow at it's the forecasted rate. That a major hiccup in the autonomous vehicles or mixed reality space has a cooling effect on computer vision.
But fundamentally the amount of data that any computer vision system needs to be able to classify objects, in reality is infinite. Because reality never stops changing, thus the amount of data one will need to accurately identify objects within it will never cease. Every quarter new products are released, styles change, packaging changes, etc meaning the synthetic data business should be a very safe place to be.
Incubation/Acceleration programs accomplishment
XRC Labs, The RLab, Intellectual Ventures New Ventures Program, Plug 'N Play's Brand & Retail Cohort.
Won the competition and other awards
We were one of Verizon's winners of their Built on 5G Challenge, for which we received $250k in non-dilutive prize money.
In 2019 won the pitch competitions at Augmented World Expo, GS1 Connect, and Data Con LA. In addition to being the winner of O'Reilly Media's People's Choice Award at their San Jose Artificial Intelligence Conference.
We were nominated for an Interactive Innovation Award in the cancelled 2020 SXSW.
We have an exclusive license to 38 patents with enforcement rights on spatial object recognition invented by Bill Gates and Satya Nadella, licensed from Intellectual Ventures.
These patent families are generally about scanning an environment and picking up on automatically executable tasks from “tags” in the environment. They could be used to prevent a company from executing its tech in an automated or passive way, among other things:
Correlating User Reaction with at Least an Aspect Associated with an Augmentation of an Augmented View
Systems and Methods for Scanning a User Environment and Evaluating Data of Interest
Presenting an Augmented View in Response to Acquisition of Data Inferring User Activity
Displaying in Response to Detecting One or More User Behaviors One or More Second Augmentations That Are Based on One or More Registered First Augmentations
These cover sharing an augmented reality scene with a second user in the same space; so a Company would be violating this if they were to display the 3D model they loaded into the scene to a second user, among other things:
Formatting of One or More Persistent Augmentations in an Augmented View in Response to Multiple Input Factors
Systems and Methods for Sharing Augmentation Data
This patent family covers executing a search in AR by pointing an AR device or limb at an item of real-world content, we can block a company in the act of actually executing the search, among other things.