How2Sign: A Large-scale Multimodal Dataset for Continuous American Sign Language

TL;DR

How2Sign introduces a comprehensive multimodal dataset for American Sign Language, enabling advancements in recognition, translation, and production by providing extensive annotated videos with multiple modalities including speech, transcripts, and 3D pose data.

Contribution

The paper presents a large-scale, multimodal ASL dataset with over 80 hours of videos and detailed 3D pose annotations, facilitating research in sign language understanding.

Findings

Synthesized videos from the dataset are understandable by ASL signers.

The dataset enables new research avenues in sign language recognition and translation.

Insights into challenges for computer vision in sign language applications.

Abstract

One of the factors that have hindered progress in the areas of sign language recognition, translation, and production is the absence of large annotated datasets. Towards this end, we introduce How2Sign, a multimodal and multiview continuous American Sign Language (ASL) dataset, consisting of a parallel corpus of more than 80 hours of sign language videos and a set of corresponding modalities including speech, English transcripts, and depth. A three-hour subset was further recorded in the Panoptic studio enabling detailed 3D pose estimation. To evaluate the potential of How2Sign for real-world impact, we conduct a study with ASL signers and show that synthesized videos using our dataset can indeed be understood. The study further gives insights on challenges that computer vision should address in order to make progress in this field. Dataset website: http://how2sign.github.io/