Official repo for the AAAI 2025 paper LazyDiT.
Prepare the ImageNet and download pretrained DiT checkpoints from original DiT.
Go to lazydit/lazy_learning/.
Set ImageNet path and pretrained model checkpoint path in run-lazy-learning-imagenet.sh.
Then run by sh run-lazy-learning-imagenet.sh.
@article{shen2024lazydit,
title={Lazydit: Lazy learning for the acceleration of diffusion transformers},
author={Shen, Xuan and Song, Zhao and Zhou, Yufa and Chen, Bo and Li, Yanyu and Gong, Yifan and Zhang, Kai and Tan, Hao and Kuen, Jason and Ding, Henghui and others},
journal={arXiv preprint arXiv:2412.12444},
year={2024}
}