The OWM benchmark comprises 95 curated videos with motion annotations, sourced from Pexels under the Pexels License. It was proposed in the paper 'Envisioning the Future, One Step at a Time' and used to evaluate the MYRIAD model. The dataset was created by CompVis and last updated on 2026-04-10.
Use Cases
- Benchmarking probabilistic motion prediction methods based on the constrained motion distribution.
- Evaluating video understanding models based on manually annotated object and motion types.
- Training models for future motion forecasting based on curated video sequences.
Strengths
- Contains 95 manually annotated videos, providing a focused benchmark size.
- Motion annotations are constrained to enable specific evaluation of probabilistic methods.
- Videos are sourced from Pexels, which provides a license for use.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- CompVis
- Collection Method
- Videos obtained from Pexels and manually annotated for relevant objects and motion types.
- Time Range
- null
- Freshness
- Last updated 2026-04-10 14:40:19; freshness should be verified.
- Geography
- null