Our highly versatile
synthetic data pipeline.
Generating useful training data for AI can be tedious and resource-heavy. Relevant data has to be acquired, preselected and labeled. Our synthetic data pipeline solves this issue by automated scenario generation and physical based sensor processing. All according to our customers use case and data structures.
Our datafactory comprises a fully automated process to generate synthetic data. After parametrizing it to match our customers use case, it is able to generate large quantities of ultra realistic training data for autonomous systems.
ASSETS & SCENE DATA
In order to match the content of real world data in quality, we host an extensive library of high-quality models to populate our metaworlds:
Our automated scene generation assembles the input data into defined scenarios. This step is shaped by a plethora of tools and parameters:
Procedural algorithms and specialized AI models empower ultra-realistic, diverse and variable environments
Parametrizable lighting, environment, profile of the terrain, density of objects, trajectories, animations and materials
Variations on consecutive image sequences or a frame by frame level
OUTPUT & METADATA
Ultimately, highly relevant synthetic training data are generated. They are easily integratable into our customer’s MLOps and have pixel-accurate labels, such as:
- Semantic group segmentation
Semantic instance segmentation
2D bounding boxes
3D bounding boxes
Body part annotation
- Skeleton/Pose .json