Word-Level American Sign Language (WLASL) video dataset

WLASL数据集是由澳大利亚国立大学澳大利亚机器人视觉中心创建的一个大规模视频数据集，专注于美国手语的单词级别识别。该数据集包含超过2000个单词，由100多名手语者表演，总计21,083个视频。数据集的创建过程涉及从互联网上收集视频，并通过自动检测和手动标注确保视频与标注的正确性。WLASL数据集旨在解决现有手语数据集词汇量有限的问题，支持深度学习方法在大型场景下的性能评估，特别适用于单词级别的手语识别研究。

The WLASL dataset is a large-scale video dataset created by the Australian Centre for Robotic Vision at the Australian National University, focusing on word-level recognition of American Sign Language (ASL). This dataset contains over 2,000 words performed by more than 100 signers, with a total of 21,083 videos. The dataset's creation process involves collecting videos from the Internet, and ensuring the correctness of both the videos and their annotations through automatic detection and manual annotation. The WLASL dataset aims to address the limited vocabulary problem of existing sign language datasets, supports performance evaluation of deep learning methods in large-scale scenarios, and is particularly suitable for word-level sign language recognition research.

搜集汇总

背景与挑战

背景概述

WLASL数据集是一个大规模视频数据集，专注于美国手语的单词级别识别，包含超过2000个单词、100多名手语者和21,083个视频，旨在解决现有手语数据集词汇量有限的问题，支持深度学习在大型场景下的性能评估，适用于手语识别研究。