First-person task video
Head-mounted or wearable views showing hands, objects and task progression.
Egocentric & ego-exo data collection
Egocentric video data collection for real-world task understanding.
Capture human activity from the participant’s point of view while preserving the task context, hands, tools, objects and environment needed for AI training or evaluation.
Capability scope
What an egocentric collection program can include
Ego-exo capture
First-person video paired with fixed or operator-positioned external views when synchronized context is required.
Narrated workflows
Project-approved narration can describe actions, tools, objects or task state during recording.
Task metadata
Task ID, participant, environment, device, timestamps, category, review state and other client-defined fields.
Operating controls
What we validate before scale and delivery
Acceptance rules vary by project, so the controls below are configured against the approved specification rather than treated as universal thresholds.
Task visibilityHands, tools and relevant actions remain visible throughout the required sequence.
Camera protocolResolution, frame rate, device placement, orientation and audio settings follow the approved specification.
Environment suitabilityLighting, privacy, permissions, safety and background conditions are checked before capture.
Batch QARepresentative samples and production batches are reviewed against project-specific rejection rules.
Specification design
Decisions that materially affect egocentric dataset quality
Useful first-person data starts with a capture protocol that matches the model objective. These points should be resolved before participant onboarding or scale.
Viewpoint and task coverage
Define where the camera is worn or positioned, which parts of the hands and workspace must remain visible, and what counts as a complete task. A technically valid file can still be unusable if the critical action happens outside the required view.
Task boundaries and narration
Specify start and end conditions, interruptions, retries and whether narration is required. When narration is part of the protocol, define what should be described and how it should relate to visible actions rather than leaving workers to improvise.
Ego-exo synchronization
When external views are required, define how recordings are aligned, how missing views are handled and which camera is authoritative for task timing. Synchronization requirements should be validated during the pilot rather than assumed at delivery.
Metadata and rejection logic
Agree the task ID, session or participant fields, environment and device fields, review states and rejection reasons that must accompany each file. This makes rework traceable and prevents ambiguous batches later.
Typical pre-pilot questionsWhat must remain visible? Which actions invalidate a take? Is audio or narration required? Are multiple views synchronized? Which metadata fields are mandatory? What is the accepted handoff structure?
Delivery structure
Outputs organized around the agreed use case.
Depending on scope, delivery may include approved raw recordings, synchronized views, task metadata, quality notes, manifests and agreed naming or folder structures.
Related capabilities