opendiffusionai/laion2b-squareish

这是一个通过合并两个数据集并进行额外过滤、去重和水印移除创建的数据集。它主要用于训练，特别强调方形图像。该数据集为实际训练添加了新的标题，而不是使用原始的LAION TEXT字段。尽管数据集尚未完成，但相较于仅通过筛选laion2b-aesthetic集合得到的版本有了显著改进。同时，README中还详细说明了数据集的大小和图像尺寸。

This dataset is created by merging two other datasets with additional filtering, deduplication, and watermark removal. It is primarily intended for training with a focus on square images. The dataset includes new captions for training purposes instead of using the original LAION TEXT field. Although the dataset is not yet complete, it represents a significant improvement over a simple filtered version of the laion2b-aesthetic set. Detailed information about the dataset size and image dimensions is also provided.