Plan the path from a 3D scene to useful training data
Updated 2026-10-30
Need labeled camera frames, depth observations or variations of a rare scenario? The directory separates tools that generate scenes, simulate robot sensors, and manage dataset delivery.
Choose the workflow first, then follow the official sources to check labels and exports. This is a starting point for evaluating pipeline fit; it does not rank vendors or promise model performance.
Browse by data workflow
Compare the documented jobs of generation libraries, procedural scene code, robot simulation and managed data platforms.
Translate the output terms
Understand segmentation, depth, optical flow and export annotations before connecting a generator to training code.
Add evidence or fix a claim
Submit a tool with official documentation or identify a dated capability note that needs correcting.
Discuss a disclosed showcase
Commercial presentations and reviews have their own space, separate from editorial data-tool entries.
Three ways to enter a data pipeline
- A programmable pipeline: Replicator is an Omniverse extension for building custom synthetic data tools. [docs.omniverse.nvidia.com]
- A label-ready render workflow: BlenderProc documents RGB, depth, normal and segmentation rendering, with COCO and BOP annotation output. [github.com]
- A multi-object video generator: Kubric combines physics and rendering with annotations for segmentation, depth and optical flow. [github.com]
- A managed workflow: Rendered.ai describes generation workflows, asset management and model training within its platform. [rendered.ai]