SahilBhatane/sli

该数据集包含多个类别的图像数据，每个类别对应一个子文件夹，文件夹名称是人类可读的标签（如stop、good morning）。每个文件夹中包含RGB图像文件（如.png、.jpg、.jpeg或.bmp）。数据集可以通过多种方式填充，包括从Hugging Face同步数据、从Mendeley下载ISL短语数据、合并到类文件夹、添加自己的捕获图像以及其他语料库。训练脚本会自动处理类平衡，以防止罕见类别在训练中被忽略。图像文件被git忽略以保持仓库小型化。

The dataset contains image data for multiple classes, with each class corresponding to a subfolder whose name is a human-readable label (e.g., stop, good morning). Each folder contains RGB image files (e.g., .png, .jpg, .jpeg, or .bmp). The dataset can be populated in various ways, including syncing data from Hugging Face, downloading ISL phrase data from Mendeley, merging into class folders, adding your own captured images, and other corpora. The training script automatically handles class balance to prevent rare classes from being ignored during training. Image files are git-ignored to keep the repository small.