1 paper
Yingheng Wang, Yaoqiang Li, Yaqin Wu +4
Representation alignment improves and accelerates diffusion training by encouraging intermediate features to match a pretrained visual encoder. However, standard alignment matches…