From Patches to Graphs: Towards Image Diffusion Models with GNNs

Xavier Sumba, Yingzhen Li

Keywords: Generative Models, Graph Neural Networks, Image Diffusion Models

TL;DR: We propose DiG, a novel graph-based architecture for diffusion models that captures both local and global dependencies in image generation across multiple resolutions.

Abstract:

Diffusion models have achieved remarkable success in high-quality image generation, typically using convolutional neural networks (CNNs) or Vision Transformers (ViTs) as backbone architectures. However, CNNs may struggle with capturing long-range dependencies, while ViTs can be computationally intensive due to their attention mechanisms. We propose the Diffusion Image GNN (DiG), a novel architecture that leverages graph-based modeling within diffusion models. By representing image patches as nodes in a graph and connecting them based on spatial relationships, DiG efficiently captures both local and global dependencies and naturally handles multi-scale features.
Empirical results demonstrate that DiG achieves competitive Frechet Inception Distance (FID) scores compared to state-of-the-art methods. To our knowledge, this is the first application of graph neural networks as a backbone within diffusion models for image generation, opening new avenues for research in generative modeling.

Primary Area: learning on graphs and other geometries & topologies

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.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

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: 8864