Sourcing field guide
Head-mounted or chest-mounted egocentric video?
Compare viewpoints using the actions and evidence your model needs.
← Data Egocentric
Fiund / Resources
The resource library.
Original guides to data egocentric: source selection, collection design, annotations, evaluation, and project review.
27 guides shown
Sourcing field guide
How to specify hand visibility in first-person datasets
Turn “hands visible” into an annotation and acceptance policy.
Sourcing field guide
Why complete task sequences matter in egocentric data
Preserve preparation, transitions, and outcomes when scoping first-person recordings.
Sourcing field guide
Planning a workplace egocentric capture session
Connect process coverage to recording boundaries and practical sample review.
Sourcing field guide
Building an action vocabulary for egocentric video
Choose labels that distinguish actions without overinterpreting the view.
Sourcing field guide
Setting temporal boundaries for first-person actions
Make action starts, ends, and overlaps reviewable.
Sourcing field guide
Scoping object state changes in egocentric datasets
Request before-and-after evidence for opening, assembling, cutting, and other changes.
Sourcing field guide
Evaluating occlusion in first-person video datasets
Treat hidden hands and objects according to the model task.
Sourcing field guide
Reviewing camera motion in egocentric recordings
Assess motion against action visibility rather than polished appearance.
Sourcing field guide
Planning lighting coverage for egocentric video
Describe transitions and local visibility within the working scene.
Sourcing field guide
Specifying participant variation in egocentric collections
Focus variation on observable behavior and deployment requirements.
Sourcing field guide
Designing object and layout variation for first-person data
Make variation interpretable across objects, arrangements, and task goals.
Sourcing field guide
Should egocentric datasets include failed attempts?
Separate outcome, execution, and visibility when recording unsuccessful tasks.
Sourcing field guide
Continuous egocentric sessions or task clips?
Choose recording units that preserve the context your model will receive.
Sourcing field guide
When audio belongs in an egocentric dataset brief
Specify whether sound is evidence, context, or excluded material.
Sourcing field guide
Checking timestamps in egocentric video deliveries
Keep clips, labels, and optional sensor streams on a documented timeline.
Sourcing field guide
What belongs in an egocentric dataset manifest?
Make sessions, clips, tasks, and capture configurations traceable.
Sourcing field guide
Avoiding duplicate clips and session leakage
Review source relationships before splitting egocentric footage.
Sourcing field guide
Testing first-person perception in unseen environments
Define what “new environment” means before requesting evaluation footage.
Sourcing field guide
How to review an egocentric capture pilot
Use a small representative review to resolve assumptions before expanding.
Sourcing field guide
Setting recording boundaries for first-person capture
Plan excluded content and review responsibilities before a camera moves through a scene.
Sourcing field guide
Existing first-person footage or a new capture project?
Compare available context, capture control, and reviewable source documentation.
Sourcing field guide
Why footage hours do not describe egocentric coverage
Compare activity, environment, and interaction coverage alongside duration.
Sourcing field guide
Instructions, narrations, and descriptions in egocentric data
Keep task intent separate from what the video actually shows.
Sourcing field guide
Combining egocentric and external camera views
Specify whether views are paired, synchronized, and useful for the same task.
Sourcing field guide
How to scope an egocentric data collection
A practical brief for first-person tasks, camera placement, coverage, and review.
Sourcing field guide
An egocentric video dataset evaluation checklist
Review viewpoint, task completeness, metadata, and permitted use before selecting first-person data.