EgoDemo is an ungated trial pack sampled from the larger EgoSuite family to let researchers inspect modalities, annotation schemas, and delivery formats without requesting full access. It exposes synchronized egocentric RGB plus frame-level 3D hand and optional full-body pose and temporal semantics, making it useful for prototyping imitation-learning and perception-to-action pipelines.
What Sets It Apart
- Compact, inspection-ready subset: contains a small number of hours (sampled 20–50h typical) so you can validate data quality and schema before applying for the full gated release — useful for quick end-to-end checks.
- Multimodal, pose-centric episodes: synchronized head/wrist RGB with left/right 3D hand pose and optional full-body pose, plus semantic event annotations — so developers can build and test grasping, contact, and manipulation workflows with realistic human-perspective data.
- Delivery-ready formats: episodes packaged as Parquet and supported multimodal bundles (LeRobot, MCAP), including frame-quality metadata where available — so integration with training pipelines and verification tools is straightforward.
Who it's for and trade-offs
Great fit if you need a realistic, annotated egocentric sample to prototype imitation learning, dexterous manipulation perception, or embodied-AI data pipelines without the overhead of acquiring the full EgoSuite releases. Look elsewhere if your project requires hundreds or thousands of hours for large-scale pretraining, or if you need wrist-camera coverage and full dataset quotas beyond the demo sample — the demo is intentionally small and intended for schema inspection and early validation.
Where it fits
Use EgoDemo to validate annotation conventions, test loader code (Parquet/MCAP/LeRobot), and run small-scale experiments or demos. For large-scale training and coverage across more tasks/scenes, consider the broader EgoSuite SKUs (EgoStandard, EgoPro, etc.) which provide the planned tens of thousands of hours and additional variants.
