How2Sign: American Sign Language Video Dataset with English Translations

How2Sign provides video data for American Sign Language (ASL) recognition. The dataset contains frontal and side-view RGB videos, with English translation text available for the frontal view. It was uploaded by aipieces and last updated on February 7, 2024.

Use Cases

• Train sign language recognition models based on frontal-view RGB video data.
• Develop multimodal translation systems based on paired video and English text data.
• Benchmark pose estimation algorithms based on available keypoints data.

Strengths

• Includes multiple data modalities: RGB video from frontal and side views, and keypoints.
• Provides English translation text for frontal-view RGB data.
• Training set is 30.7 GB, validation set is 1.65 GB, and test set is 2.24 GB, indicating substantial scale.

Limitations

• Gloss annotations are not currently available, limiting certain linguistic analyses.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

aipieces

Freshness

Last updated 2024-02-07 02:16:36