Training data for facial recognition

Data annotation adds defined labels to source data so models can learn a specific task. For facial recognition, examples include face boxes and landmarks and facial region masks. Unitlab connects this work in its data annotation platform.

Choose the tools that match the source data: image annotation, video annotation. Keep linked sources together when the task requires shared context. Confirm input formats and annotation requirements before starting a project.

This page focuses on face boxes, facial landmarks, and visible region masks. Identity & Biometrics covers the broader identity-data workflow. Annotation describes visible source content; identity labels must come from your authorized reference data.

Define classes, required properties, and boundary rules before labeling. Use examples such as face boxes and landmarks to resolve ambiguous cases. See the classes and annotation types guide.

Use data curation to inspect examples and filter available metadata. Plan coverage across poses, occlusions, illumination, and image quality, then check for missing or overrepresented conditions before annotation.

Use annotation quality assurance to inspect labels against the task guidelines. Consensus helps compare annotator agreement; Quality Gate stages apply configured checks before work advances. Route uncertain examples to the appropriate reviewer.

A dataset version records a source-data selection; an annotation release packages the annotation outputs for downstream use. Use dataset management to inspect and organize data, and consult the guide to annotation releases before preparing training exports.

Choose an output format supported for your annotation task and validate the result with your training code. Read the export formats guide and the API, SDK, and CLI documentation for automation and integration options.

Start with a representative sample, a clear label specification, and an agreed review process. Read the image annotation documentation or discuss your workflow with the Unitlab team.