Teaching Computers to See with Blender

Machine learning is a hot topic right now and everyone is looking to see how they can use their data to teach computers to understand the world. At LexSet, we use open-source technology to turn our customer's 3D content into synthetic image datasets that are used to train advanced machine vision models. We have built a scalable system for rendering highly parametric content in the cloud to enable rapid, low-cost iteration. We are using a number of existing tools built by the blender community and combine them into a powerful machine learning toolkit. In the future, we will integrate machine learning directly into the rendering pipeline to automatically optimize the rendering parameters for specific machine learning tasks. This talk will detail our system, discuss challenges, and show what's involved in building an AI startup around Blender 2.8.


  • Principle Engineer at LexSet building web-based tools to enable cloud-scale, photorealistic, 3D rendering used to train computers to understand the visual world. Formerly worked in the 3D printing community @Thingiverse and @Shapeways. A Blender user since 2006 (v2.42!), Angelo loves all things procedural and suspects that the fundamental components of the universe may be node-based!