brl-xfact/polaris_imagereward

这个数据集包含了四个主要特征字段：refs和cand为字符串类型，img为一个三层嵌套的浮点数序列，human_score为浮点数类型。数据集分为训练集、验证集和测试集，分别包含81909、28852和28842个示例。整个数据集的大小为约525GB，下载大小约为121GB。提供了默认配置，指定了训练、验证和测试数据文件的路径。

The dataset consists of four main feature fields: refs and cand are string types, img is a three-level nested sequence of floats, and human_score is a float type. The dataset is divided into training, validation, and test sets, containing 81909, 28852, and 28842 examples respectively. The total size of the dataset is approximately 525GB, with a download size of about 121GB. A default configuration is provided, specifying the paths to the data files for training, validation, and testing.