mair-lab/earl-datasets

EARL标记化数据集包含在论文[*The Promise of RL for Autoregressive Image Editing*]中使用的所有数据集的VQ标记化版本。这些数据集使用EARL GitHub仓库的官方脚本进行标记化，将图像转换为与EARL自回归模型兼容的离散VQ令牌。每个样本包括原始和编辑图像的标记化表示、描述编辑的文本指令，以及在可用的情况下，记录编辑过程中间步骤的推理轨迹（“思考”）。该数据集集合包含简单编辑（简短、直接的指令）和复杂编辑（多步骤或抽象转换），如论文中所定义。

The EARL Tokenized Datasets contain the VQ-tokenized versions of all datasets used in the paper The Promise of RL for Autoregressive Image Editing. Tokenization was performed using the official scripts from the EARL GitHub repository, converting images into discrete VQ tokens compatible with EARLs autoregressive models. Each sample includes the tokenized representation of the original and edited images, text instructions describing the edit, and optional reasoning traces that document intermediate steps in the editing process. The collection includes both simple edits (short, direct instructions) and complex edits (multi-step or abstract transformations) as defined in the paper.