BriefGPT.xyz
Feb, 2024
文本-图像扩散与偏好的密集奖励观点对齐
A Dense Reward View on Aligning Text-to-Image Diffusion with Preference
HTML
PDF
Shentao Yang, Tianqi Chen, Mingyuan Zhou
TL;DR
通过引入时间折扣机制以适应T2I生成层次结构，本文提出了一种可行的对齐目标，强调T2I反向链中的初始步骤，实验证明该方法在单个和多个提示生成方面与相关基线方法具有竞争力。
Abstract
Aligning
text-to-image diffusion model
(T2I) with
preference
has been gaining increasing research attention. While prior works exist on directly optimizing T2I by
→