The diffusion model has long been plagued by scalability and quadratic complexity issues, especially within transformer-based structures. In this study, we aim to leverage the long sequence modeling capability of a State-Space Model called Mamba to extend its applicability to visual data generation. Firstly, we identify a critical oversight in most current Mamba-based vision methods, namely the lack of consideration for spatial continuity in the scan scheme of Mamba. Secondly, building upon this insight, we introduce a simple, plug-and-play, zero-parameter method named Zigzag Mamba, which outperforms Mamba-based baselines and demonstrates improved speed and memory utilization compared to transformer-based baselines. Lastly, we integrate Zigzag Mamba with the Stochastic Interpolant framework to investigate the scalability of the model on large-resolution visual datasets, such as FacesHQ 1024×1024 and UCF101, MultiModal-CelebA-HQ, and MS COCO 256×256 . Code will be released at https://taohu.me/zigma/
@article{arxiv.2403.13802,
title = {ZigMa: A DiT-style Zigzag Mamba Diffusion Model},
author = {Vincent Tao Hu and Stefan Andreas Baumann and Ming Gui and Olga Grebenkova and Pingchuan Ma and Johannes Schusterbauer and Björn Ommer},
journal= {arXiv preprint arXiv:2403.13802},
year = {2024}
}