We present ImageReward -- the first general-purpose text-to-image human
preference reward model -- to address various prevalent issues in generative
models and align them with human values and preferences. Its training is based
on our systematic annotation pipeline that covers both the rating and ranking
components, collecting a dataset of 137k expert comparisons to date. In human
evaluation, ImageReward outperforms existing scoring methods (e.g., CLIP by
38.6\%), making it a promising automatic metric for evaluating and improving
text-to-image synthesis. The reward model is publicly available via the
\texttt{image-reward} package at https://github.com/THUDM/ImageReward.

本文介绍了 ImageReward，第一个通用的文本到图像人类偏好奖励模型，通过系统注释流程，其训练基于涵盖评级和排名组件的数据集，包括 137k 个专家比较。在人类评估中，ImageReward 超越了现有的评分方法（例如 CLIP），成为评估和改进文本到图像合成的有前途的自动度量标准。该奖励模型可通过 	exttt {image-reward} 包在 https://github.com/THUDM/ImageReward 上公开获得。