VIDEOEGOCENTRIC
First-Person Activity Clips
Egocentric activity video for models that need to understand actions, objects and task progression from the actor's point of view.
View datasetVIDEOENVIRONMENT DIVERSITY
Multi-Environment Activity Clips
A model that works in one setting may behave differently when the background, lighting, viewpoint or surrounding environment changes. Multi-Environment Activity Clips keep those conditions explicit so environment variation becomes part of the dataset design rather than hidden noise.
The same or related activities can be captured across a planned set of contexts, with condition metadata attached to the relevant clips or segments. This supports model work where the relationship between activity and environment needs to be understood rather than averaged away.
Technical structure
- Activity identifiers shared across selected environments where appropriate
- Environment, lighting, viewpoint or other condition metadata
- Temporal activity segments where the task needs localised examples
- Coverage information aligned to the planned collection matrix
Typical applications
Robustness evaluation, domain generalisation, activity recognition, context-sensitive modelling and held-out condition analysis.
Delivery note
The environment matrix is defined during scoping, including any grouping needed for the intended evaluation design.
Delivery
The exact package depends on the collection scope. Where relevant, delivery can include task-specific records, manifests, provenance fields, stable identifiers, SHA-256 hashes, MLCommons Croissant 1.0 metadata and loading instructions.