LexSet Synthetic Data Engine


For computer vision ML engineers who are tired of the long and tedious process of sourcing and annotating found photographs to train their AIs, LexSet has a better solution to supply training data. 


Instead of using human-in-the-loop annotators at mechanical-Turk click farms, which is slow and costly; LexSet’s synthetic data engine can create perfectly annotated images of whatever objects or environments are needed, for a fraction of the price.


Our platform utilizes our immense library of 3D CAD files which we procedurally place into massively parallel, algorithmically generated simulation environments. We’ve solved the “over-fitting problem” that has traditionally hindered synthetic data from being able to replace photographs, by building our own rendering pipeline that produces what we call “physically accurate lighting”. 

LexSet Annotations


Tailored to your specifications, our training images come with 100% accurate annotations which include:

  • Object Labels

  • 2D bounding boxes

  • 3D bounding boxes

  • Depth maps

  • Texture maps

  • Semantic Segmentation masks

  • Camera intrinsics , including position, orientation, and lens type


We specialize in interior environments including home & office, retail, warehouse, manufacturing facility and more. We also make data for object detection and can create data for any object you need to recognize.

LexSet Product Road Map

  1. Self-service object recognition data [coming Q1, 2020]

    • Upload CAD (or choose from our library), create your own training data


  2. World creator [coming Q4, 2020]

    • Multiple CAD Objects procedurally loaded into an environment

    • Define rules for the object placement

    • Generate environments from residential interiors, to offices, to warehouses, to hospitals


Tel: 908.793.8153

LexSet c/o New Lab

19 Morris Ave, Building 128

Brooklyn, NY 11205

  • LinkedIn - White Circle
  • Facebook - White Circle
  • noun_Email_1198069_edited_edited_edited.

© 2017 by LexSet.