A directory for the training-data handoff

Updated 2026-10-12

This directory is for computer vision and robotics teams working out how to create labeled observations from controllable scenes. It covers generation code, robot sensor simulation, and commercial dataset workflows.

Start with the labels and sensor output your training code expects. Then check scene control, export formats, compute requirements and dataset rights in the official documentation. A rendered image alone does not establish that a dataset will improve a model.

CUVISS edits the directory. Inclusion is free, evidence is linked, and paid partnerships appear in a separate section. We do not run a comparative benchmark or certify a provider.

Evidence and dataset fit

• A task defines the category. We group entries by the data workflow they document, with research generators identified as code projects rather than managed services.
• Claims keep their sources. Each capability note links to official documentation, a project repository or a provider page and records when it was checked.
• Output quality needs your test. The directory does not infer training gains, label correctness or a closed sim-to-real gap from a vendor demonstration.
• Free editorial consideration. Submissions and corrections are reviewed without a listing fee. A commercial arrangement cannot buy inclusion or rewrite an editorial card.
• Disclosed commercial space. Sponsored showcases and any compensated reviews are identified separately from the data-tool directory.
• Describe a reproducible handoff. Useful submissions explain sensor modalities, label types, export formats and access requirements. Keep private datasets and credentials out of the form.