Lune

CVPR2025

Diffusion Model is Effectively Its Own Teacher

Xinyin Ma, Runpeng Yu, Songhua Liu, Gongfan Fang, Xinchao Wang

DOI出版方

摘要

DiT-XL/2 For ImageNet 512×512 (b) Stable Diffusion v1.5 Figure 1. Generated Images before (left) and after (right) our proposed self-distillation framework on DiT-XL/2 for ImageNet 512⇥512 and Stable Diffusion v1.5 with 10 DDIM sampling steps.