Meaningful Pose-Based Sign Language Evaluation
Año:
2025
Editorial:
ArXiv
Tipo de código:
Soporte:
Temas
Medios de comunicación y acceso a la información » Tecnologías
Detalles
We present a comprehensive study on meaningfully evaluating sign language utterances in
the form of human skeletal poses. The study covers keypoint distance-based, embeddingbased, and back-translation-based metrics. We show tradeoffs between different metrics in different scenarios through (1) automatic meta-evaluation of sign-level retrieval,
and (2) a human correlation study of textto-pose translation across different sign languages. Our findings, along with the opensource pose-evaluation toolkit, provide a practical and reproducible approach for developing and evaluating sign language translation
or generation systems.
