Industry Event
Vision Tech AI Session: Beyond the Real World: Building High-Fidelity Industrial Vision with Synthetic Images
21 January 2026 • 2:00 PM - 2:30 PM PST | West Expo Stage 2 (Moscone West, Exhibit Level)
The performance of deep learning models for industrial machine vision is fundamentally constrained by the quality and breadth of training data. While modern imaging systems can capture immense detail, acquiring comprehensive datasets for industrial inspection remains a critical bottleneck. Real-world data collection is often expensive, time-consuming, and fails to capture the full distribution of rare but critical corner cases, such as subtle process variations, novel defect classes, or challenging lighting conditions. This data gap severely limits the robustness and reliability of automated vision systems in production environments.
This session presents a methodology for overcoming this challenge by augmenting and, in some cases, replacing real-world images with physically-accurate synthetic data. We will move beyond basic photorealism and delve into the technical specifics of generating high-fidelity, synthetic datasets tailored for demanding industrial applications.
Topics will include techniques for simulating material properties, modeling sensor noise and lens distortion, and applying domain randomization to ensure models generalize from simulation to reality. The session will provide a technical roadmap for how to leverage simulation to create perfectly-labeled, pixel-perfect ground truth data at scale, enabling the development of next-generation industrial vision systems.
Brian Geisel is the CEO of Symage, a synthetic data company using physics-based simulation to generate high-fidelity, labeled datasets that close the sim-to-real gap in machine learning.
Drawing on decades of technical expertise in robotics and AI systems, Brian founded Symage to address a critical challenge: real-world data scarcity. The platform produces physics-accurate, photorealistic datasets that mirror real-world conditions with precision, enabling more effective model training while eliminating the cost and complexity of field data collection.
With Symage, Brian is proving that engineered data can replace expensive field work for production AI systems.
MENU: Coffee, decaf, and tea will be available nearby.
SETUP: Theater seating.
This session presents a methodology for overcoming this challenge by augmenting and, in some cases, replacing real-world images with physically-accurate synthetic data. We will move beyond basic photorealism and delve into the technical specifics of generating high-fidelity, synthetic datasets tailored for demanding industrial applications.
Topics will include techniques for simulating material properties, modeling sensor noise and lens distortion, and applying domain randomization to ensure models generalize from simulation to reality. The session will provide a technical roadmap for how to leverage simulation to create perfectly-labeled, pixel-perfect ground truth data at scale, enabling the development of next-generation industrial vision systems.
Speaker
 |
|
Brian Geisel is the CEO of Symage, a synthetic data company using physics-based simulation to generate high-fidelity, labeled datasets that close the sim-to-real gap in machine learning.
Drawing on decades of technical expertise in robotics and AI systems, Brian founded Symage to address a critical challenge: real-world data scarcity. The platform produces physics-accurate, photorealistic datasets that mirror real-world conditions with precision, enabling more effective model training while eliminating the cost and complexity of field data collection.
With Symage, Brian is proving that engineered data can replace expensive field work for production AI systems.
Event Details
FORMAT: Oral presentation followed by audience Q&A.MENU: Coffee, decaf, and tea will be available nearby.
SETUP: Theater seating.