Sandwich: Bidirectional Mapping between Action Physical-Semantic Space
What it is
Sandwich provides a unified framework for human activity understanding by establishing a structured semantic space for action labels. It includes Pangea, a large-scale multi-modal action database, and a bidirectional mapping system to align physical actions with semantic meanings.
The problem it solves
The lack of a unified semantic space for action labels across different datasets, which complicates multi-modal action recognition and generation.
What you could use it for
- Aggregating multiple heterogeneous action datasets into a single unified label system.
- Training large-scale models on diverse human activities using a consistent semantic hierarchy.
- Learning bidirectional mappings between human action physical states and semantic labels.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.