Heterogeneous Image GNN: Graph-Conditioned Diffusion for Image Synthesis
This addresses the challenge of handling diverse, relational conditioning variables in image generation for computer vision applications, representing an incremental advance over existing conditioning methods.
The paper tackles the problem of conditioning diffusion-based image synthesis models on complex relational data by introducing Heterogeneous Image Graphs (HIG), a novel representation that models conditioning variables and target images as interconnected graphs. The method improves state-of-the-art performance on COCO-stuff and Visual Genome datasets with various conditioning inputs.
We introduce a novel method for conditioning diffusion-based image synthesis models with heterogeneous graph data. Existing approaches typically incorporate conditioning variables directly into model architectures, either through cross-attention layers that attend to text latents or image concatenation that spatially restrict generation. However, these methods struggle to handle complex scenarios involving diverse, relational conditioning variables, which are more naturally represented as unstructured graphs. This paper presents Heterogeneous Image Graphs (HIG), a novel representation that models conditioning variables and target images as two interconnected graphs, enabling efficient handling of variable-length conditioning inputs and their relationships. We also propose a magnitude-preserving GNN that integrates the HIG into the existing EDM2 diffusion model using a ControlNet approach. Our approach improves upon the SOTA on a variety of conditioning inputs for the COCO-stuff and Visual Genome datasets, and showcases the ability to condition on graph attributes and relationships represented by edges in the HIG.