top of page

Quality Control 

Generating synthetic data for quality control involves creating virtual models that accurately replicate the physical and visual characteristics of real-world items, allowing for the production of diverse datasets under controlled conditions. This method is invaluable for training machine learning algorithms, as it provides ample and varied data without the constraints and costs associated with real-world data collection.

Physics-based Modeling

Physics-based modeling of assortments involves using physical laws and principles to simulate the behavior and interaction of a collection of objects under various conditions. This approach can accurately predict phenomena such as the stacking, collision, and spatial distribution of items, enhancing realism and efficiency in simulations and real-world applications.

Object Categories

Geometric modeling of objects like seeds, screws, or chocolate involves creating detailed, mathematically-defined representations of their shapes and sizes, which can vary significantly from one object to another. This method allows for the precise simulation of how these objects interact in an assortment, such as fitting together in packaging or orienting themselves in a container, facilitating efficient design and arrangement

Visual Fidelity

Rendering object assortments with high visual fidelity involves using advanced graphics techniques to realistically depict the textures, lighting, and shadows of grouped items, enhancing their appearance and spatial relationships. This approach not only improves the visual accuracy and appeal of the scenes but also aids in better understanding the physical interactions and aesthetics of diverse object collections in virtual environments.

Custom Annotations

Custom annotations like instance labels or semantic segmentation masks can be generated for training data to provide detailed information about each object within an image, distinguishing individual items or categorizing them into specific classes. These annotations are essential for training machine learning models to accurately recognize and understand the content of images, enabling applications in computer vision such as autonomous driving, medical imaging, and object detection.

Automatic Defect Detection

image12 (2).jpg

Neural networks trained on synthetic data can effectively identify and classify defects with high precision, enhancing quality control processes across manufacturing and production industries. Data generation involves creating detailed synthetic datasets that simulate various imperfection types and severities, providing invaluable resources for training machine learning models to accurately detect and classify welding flaws.

Welding Defects

Paint Cracks/Splatter


Object Adhesion 



Complex Configurations

This is your Team section. Briefly introduce the team then add their bios below. Click here to edit.

Controllable Data Variance

We leverage procedural modeling for generating varied data sets. With procedural modeling we can  systematically vary parameter configurations to investigate how changes of these parameters affect the output.


Modeling of Pest Traps

Modeling 3D scenes for traps targeting pests like bugs or moths involves generating detailed simulations to optimize the placement and design of traps and bugs for maximum effectiveness for data generation.

image1 (2).jpg
image3 (2).jpg

Food and Fruit Assortments


An assortment of vine grapes typically showcases a variety of grape clusters, each differing in size, color, and grape density.


Physics-based Simulation meets Procedural Modeling

Any Assortment

Procedural modeling combined with physics-based simulation enables the creation of any assortment of objects. The fusion of these techologies is particularly useful to model how objects behave under different conditions, be it in stacking, collision, or spatial arrangement.

bottom of page