Better Data
for Better Vision
* turn up volume for full description
"Building high-performance AI models requires high-quality data, which can be hard to come by. LexSet’s synthetic data technology solves this challenge by creating custom computer vision training datasets using 3D models simulated in a variety of conditions. LexSet is working with customers to enable autonomous machines, starting with indoor environments." - Roger Chen, O'Reily Media
CV without photos via
Training Data as a Service
To enable better computer vision development, LexSet has created TDaaS (Training Data as a Service), using 3D content to create photo-realistic synthetic 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 robotic object recognition and navigation, without needing photos in the training set.
Photorealistic
Pre-annotated Rendered
Environments
Segmentation Masks
Simulated
Depth Maps
Top-down
Environment Maps
Other's have tried to use generated synthetic training data to power computer vision systems, but they ran into the "over-fitting problem", where they taught an AI how to recognize a rendering and not real life. At LexSet, we have solved this by building our own "physically accurate" rendering engine, to accurately simulate real-world lighting conditions.
Benefits of Synthetic Data
USE CASES
Domestic
& Service Robots
Retail Inventory & Restocking
For service robotics like robot vacuum cleaners and the like; which need to improve obstacle avoidance, spatial navigation, course mapping, and object recognition- LexSet can simulate the robot’s camera and then generate synthetic image data from an infinite number of variable residential and office environments. Imagine being able to say “Hey Alexa, vacuum in front of the couch.”
For retail environments where inventory tracking is being managed by robots, handheld tablets, or overhead camera systems; LexSet can use 3D models of CPG goods to create vast quantities of synthetic image data. Improving the accuracy and precision of counting inventory, creating restocking alerts, enabling cashierless checkout, and most importantly maintaining planogram compliance- making sure products are in the correct shelf position.
Supply Chain & Inspection Drones
Using 3D scans and manufacturing CAD of existing parts and assemblies; LexSet can generate synthetic image data for a variety of very useful applications in industrial, warehouse, and manufacturing environments. Enabling drones and CCTV camera systems to keep realtime counts of inventory, as well as enabling automation of QA inspection. This also has many applications for industrial safety and maintenance.
Warehouse & Factory Logistics
Proper part, product, and tool identification is an issue for every factory and warehouse. For fulfillment and logistics where technicians and robots alike need to identify things and get them where they need to go. Useful for everything from routing items for fulfillment; to aiding a technician who needs to know where to find a replacement part, or where to put away a given tool.
Augmented &
Mixed Reality
Apps
For AR apps on mobile devices and XR headsets, LexSet improves object recognition and the overall interactivity of an app to the environment around them, enhancing spatial awareness. Imagine if the holographic characters in a mixed reality game could interact with the actual furniture inside your home, what if they could sit in the chair across from you, or pick up holographic objects off the coffee table.
.png){kind=logo}
