leekezar / SemLex

[This repo will be online very soon!] The Sem-Lex Benchmark contains 84k isolated American Sign Language signs from a vocabulary of 3,149. This repo contains instructions for downloading the data and replicating the evaluation in our ASSETS '23 paper (link below).

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

The Sem-Lex Benchmark: Modeling ASL Signs and Their Phonemes

The Sem-Lex Benchmark contains 91,148 isolated sign videos representing a vocabulary of 3,149 American Sign Language signs. 41 deaf participants contributed the videos, which were aligned with ASL-LEX 2.0 and ASL SignBank by a team of linguists. The Benchmark may be combined with ASL Citizen for a total of 174k videos. Read the paper here.

A Sankey diagram illustrates different partitions of the benchmark dataset.

Download the Benchmark & Pretrained Models

The videos are available for public download. Please read and agree to the terms of use to access the download links. Pose estimation files for each video will be available soon.

Pretrained SL-GCN models are available for gloss recognition, phoneme recognition, and phoneme+gloss recognition (most accurate at ISR).

Replicate the Experiments

Details coming soon!

About

[This repo will be online very soon!] The Sem-Lex Benchmark contains 84k isolated American Sign Language signs from a vocabulary of 3,149. This repo contains instructions for downloading the data and replicating the evaluation in our ASSETS '23 paper (link below).

License:Apache License 2.0