Per-object 3D manipulation extractions from the EPIC-KITCHENS-100 dataset. The data was used to train ObjectForesight, a 3D object-centric dynamics model that predicts future 6-DoF object poses and trajectories from short egocentric video. The dataset was authored by raivn and last updated on June 6, 2026.
Use Cases
- Training 3D object-centric dynamics models based on extracted 6-DoF object poses.
- Predicting future object trajectories based on short egocentric video sequences.
- Benchmarking manipulation forecasting algorithms based on per-object 3D extractions.
Strengths
- Derived from the established EPIC-KITCHENS-100 dataset.
- Provides raw 3D object extractions, not pre-windowed trajectories.
- Associated with a published paper (arXiv:2601.05237) and a trained model.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count and file formats are unknown, which may limit suitability assessment.
- Data may reflect the temporal and activity bias inherent to the source EPIC-KITCHENS-100 dataset.
Provenance
- Source
- EPIC-KITCHENS-100
- Collection Method
- Extracted via a curation pipeline (RustinS/ObjectForesight-Data).
- Freshness
- Last updated 2026-06-06 10:52:11; freshness should be verified.