Warehouse + Logistics
Diverse consumer goods, logistics packaging and other warehouse items are critical for automation systems to be able to reason about. SBX has built industrial vision datasets in a variety of applications, accelerating critical perception capabilities.
Computer vision has many tasks related to detecting, tracking and visually reasoning about people. SBX is capable of generating diverse scenes for digital humans. SBX synthetic data is a great way to build vision tasks focused on human faces and bodies.
Organics + Produce
Food processing and agriculture are exciting areas of increasing automation. Challenges in building robust vision systems arise when the items deform, have wide appearance variations, and are hard to manually annotate. SBX has created synthetic datasets that help food + agriculture automation systems detect and process their items of interest.
Construction + Housing
Modern housing and construction automation is increasingly relying on computer vision. Lots of diversity in materials, scenes and structures makes collecting and annotating vision datasets a challenge. SBX generates diverse training data to match the variety items and materials in housing and construction applications
Manufacturing + Defects
Visual defect inspection systems are important in many manufacturing processes. Challenges arise when looking for data that includes physical defects and item damage. SBX can create vision datasets with simulated defects that can be used to train real-world detectors
Vehicles are important objects to reason about in many applications. From autonomy systems to visual inspection, capturing and annotating diverse vehicle data can be time-consuming and costly. SBX can create diverse vehicles and scenes.
Text + Documents
Text recognition and document processing is an important problems in computer vision. It can be quite challenging to create well-annotated datasets, with a wide range of print quality and defects. SBX has built synthetic text datasets that help clients minimze the costly and slow annotation process of using real data alone.