Rethinking cluster-conditioned diffusion models

arXiv:2403.00570v1 Announce Type: cross
Abstract: We present a comprehensive experimental study on image-level conditioning for diffusion models using cluster assignments. We elucidate how individual components regarding image clustering impact image synthesis across three datasets. By combining recent advancements from image clustering and diffusion models, we show that, given the optimal cluster granularity with respect to image synthesis (visual groups), cluster-conditioning can achieve state-of-the-art FID (i.e. 1.67, 2.17 on CIFAR10 and CIFAR100 respectively), while attaining a strong training sample efficiency. Finally, …

abstract arxiv cluster clustering components cs.ai cs.cv cs.lg datasets diffusion diffusion models experimental image impact show study synthesis type