Dataset Management and Versioning for AI Teams

Dataset management helps AI teams turn already annotated data into organized, traceable training inputs. In Unitlab, teams can visually inspect examples, filter relevant data, manage versions, and prepare reviewed outputs for training. It connects the results of data annotation with a repeatable delivery workflow.

Data curation focuses on finding and selecting relevant source data. Dataset management focuses on organizing, visually checking, filtering, and versioning annotated data for downstream use. Together, they connect deliberate data selection with traceable training-data delivery.

Yes. Inspect the underlying data and annotations, then use available filters and properties to narrow the examples you need to check. Resolve labeling issues through quality assurance, review, and rework before preparing a training release.

A dataset version preserves the source-data membership of a dataset at a specific point. An annotation release freezes a project's annotation output for delivery. Use dataset versions to identify your input data and project releases to identify the labeled output used for training.

Yes. A published dataset version can provide source data for another project. The project works with its own copy, so later annotation changes do not modify the source dataset. See dataset versions and history for the workflow.

Unitlab supports workflows for images, video, audio, text, documents and PDFs, HTML webpages, medical imaging, whole-slide pathology, geospatial imagery, sensor and time-series data, and tabular data. Explore multimodal annotation when related data needs shared context, and choose the appropriate output format for each training workflow.

Inspect the examples and resolve outstanding issues through your annotation quality workflow. Record the source dataset version, create a project release of the reviewed annotations, and export in a supported format suited to your training pipeline. This makes the input data and delivered labels easier to trace.

Use the API documentation, Python SDK guide, and CLI documentation to identify supported integration steps. Keep dataset-version and release identifiers alongside training runs so your team can track which data and annotations were used.