Synthetic Data 2026
An afternoon on the next generation of vision models for physical AI, trained on synthetic data for the rare, safety-critical cases nobody can capture for real. Keynotes from Synthgen and Demcon.
The next generation of vision models will learn from cases nobody could collect
The cases that matter most to a vision system, a hairline crack, a misaligned component, a diseased plant, are the ones nobody can capture at scale. Waiting months for a few hundred real examples is how vision projects stall, while the rare case slips through at the full volume of the line.
- Date 20 October 2026
- Time 16:00 to 18:00
- Location AI Innovation Center, HTC 5
- Organization Synthgen
- Language English
The next generation of vision models is trained more and more on synthetic data: rare, safety-critical cases simulated with pixel-perfect labels, learned before they ever happen for real. This afternoon is about that shift, where it already works, where it does not yet, and what it means for anyone running cameras on a line, in a field or in a clinic.
Keynotes from Synthgen and Demcon, and the conversation after
A keynote from Synthgen on training vision systems on defects you have never seen: why projects stall on data, how a rare case is modelled and generated with the Synthgen Engine, how the gap between synthetic and real is closed, and how a team checks that a model trained on synthetic data holds up on its own line.
Then Marnix Zoutenbier and Vincent Bos of Demcon Data Driven Solutions on what synthetic data makes possible for computer vision and physical AI, with examples from their own projects. Their team is the data science and machine learning design house of the Enschede high-tech group Demcon, and makes synthetic data for computer vision as a service.
For quality managers, vision and ML engineers and process engineers in precision machining, semiconductor and electronics, agri-robotics, NDT and medical devices. You leave with a practical test for whether your vision problem is a data problem.