Unveiling the Secret of AdaLN-Zero in Diffusion Transformer
Keywords: Diffusion transformer, zero-initialization, image generation
Abstract: Diffusion transformer (DiT), a rapidly emerging architecture for image generation, has gained much attention. However, despite ongoing efforts to improve its performance, the understanding of DiT remains superficial. In this work, we delve into and investigate a critical conditioning mechanism within DiT, adaLN-Zero, which achieves superior performance compared to adaLN. Our work studies three potential elements driving this performance, including an SE-like structure, zero-initialization, and a “gradual” update order, among which zero-initialization is proved to be the most influential. Building on this insight, we heuristically leverage Gaussian distributions to initialize each condition modulation, termed adaLN-Gaussian, leading to more stable and effective training. Extensive experiments following DiT on ImageNet1K demonstrate the effectiveness and generalization of adaLN-Gaussian, e.g., a notable improvement of 2.16% in FID score over adaLN-Zero.
Primary Area: generative models
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3334
Loading