andropar/relaion2b-natural

relaion2B-en-research-safe自然度分数数据集，基于ReLAION-2B-en-research-safe数据集，包含超过200万张图片的URL及其自然度分数，分数范围从0到1，分数越高表示图片越自然。数据集用于过滤自然/照片内容、计算机视觉训练数据的质量评估、图像自然性的研究以及训练视觉模型之前的数据预处理。

relaion2B-en-research-safe natural scores dataset, based on the ReLAION-2B-en-research-safe dataset, contains over 2 million image URLs with their naturalness scores, ranging from 0 to 1, with higher scores indicating more natural images. The dataset is used for filtering natural/photo content, quality assessment of computer vision training data, research on image naturalness, and as a preprocessing step before training vision models.