CSL-Web482: A web-collected Chinese sign language image dataset of 482 isolated words（32175）

Chinese Sign Language is an important communication tool for people with hearing impairments, but the development of automatic sign language recognition systems is limited by the lack of high-quality datasets. Existing Chinese sign language datasets are mostly collected in laboratory settings, with uniform backgrounds and a limited number of presenters, making it difficult to reflect the complexity of real-world applications. Sign language recognition systems deployed in practice need to handle diverse backgrounds, different performance styles, and complex lighting conditions. To fill this gap, we collected and constructed the CSL-Web482 dataset. Based on the "Chinese Sign Language" standard manual, this dataset systematically collects sign language images for 482 commonly used isolated words, all extracted from publicly available online videos, retaining the diversity and complexity of the real world. This dataset aims to provide authentic, diverse, and standardized image data for research in Chinese sign language recognition, promoting the development of robust sign language recognition algorithms.