English

LipsFormer: Introducing Lipschitz Continuity to Vision Transformers

Computer Vision and Pattern Recognition 2023-04-20 v1 Artificial Intelligence Machine Learning

Abstract

We present a Lipschitz continuous Transformer, called LipsFormer, to pursue training stability both theoretically and empirically for Transformer-based models. In contrast to previous practical tricks that address training instability by learning rate warmup, layer normalization, attention formulation, and weight initialization, we show that Lipschitz continuity is a more essential property to ensure training stability. In LipsFormer, we replace unstable Transformer component modules with Lipschitz continuous counterparts: CenterNorm instead of LayerNorm, spectral initialization instead of Xavier initialization, scaled cosine similarity attention instead of dot-product attention, and weighted residual shortcut. We prove that these introduced modules are Lipschitz continuous and derive an upper bound on the Lipschitz constant of LipsFormer. Our experiments show that LipsFormer allows stable training of deep Transformer architectures without the need of careful learning rate tuning such as warmup, yielding a faster convergence and better generalization. As a result, on the ImageNet 1K dataset, LipsFormer-Swin-Tiny based on Swin Transformer training for 300 epochs can obtain 82.7\% without any learning rate warmup. Moreover, LipsFormer-CSwin-Tiny, based on CSwin, training for 300 epochs achieves a top-1 accuracy of 83.5\% with 4.7G FLOPs and 24M parameters. The code will be released at \url{https://github.com/IDEA-Research/LipsFormer}.

Keywords

Cite

@article{arxiv.2304.09856,
  title  = {LipsFormer: Introducing Lipschitz Continuity to Vision Transformers},
  author = {Xianbiao Qi and Jianan Wang and Yihao Chen and Yukai Shi and Lei Zhang},
  journal= {arXiv preprint arXiv:2304.09856},
  year   = {2023}
}

Comments

To appear in ICLR 2023, our code will be public at https://github.com/IDEA-Research/LipsFormer

R2 v1 2026-06-28T10:11:29.034Z