Aggregating MediaPipe Holistic landmark features extracted from the How2Sign American Sign Language (ASL) video corpus. It provides synchronized skeletal and facial keypoints across three categories: body pose, hand gestures, and facial expressions.
Use Cases
- Train sign language recognition models using the extracted hand and pose landmark coordinates.
- Develop facial expression analysis tools for ASL non-manual markers using the 468 facial landmark points.
- Build gesture-to-text translation systems by mapping sequences of holistic landmarks to gloss or English text.
Strengths
- Features extracted using the MediaPipe Holistic pipeline for synchronized body, hand, and face tracking.
- Derived from the How2Sign dataset, a large-scale multimodal collection of American Sign Language.
- Includes landmark coordinates for 3D pose, 21 hand points per hand, and 468 facial landmarks.