How2Sign Processed Common 450: Keypoint Features for Sign Language Recognition

SharoonArshad processed the How2Sign dataset into a ZIP archive, last updated on 2026-05-17. The data consists of sequences of 450 features per frame, derived from holistic keypoints. The processed format combines 225 position and 225 velocity features from 75 body landmarks.

Use Cases

• Training sign language recognition models based on processed keypoint sequences.
• Developing human motion analysis algorithms based on combined position and velocity features.
• Benchmarking pose estimation models on a standard sign language dataset.
• Researching gesture recognition based on holistic body landmarks.

Strengths

• Features are processed into a consistent format of 450 dimensions per frame.
• Data combines position and velocity features for each of 75 body landmarks, which may aid motion analysis.

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

huggingface

Collection Method

Processed from the original How2Sign holistic keypoints dataset.

Freshness

Last updated 2026-05-17 20:47:56; freshness should be verified.