SEGA: A Stepwise Evolution Paradigm for Content-Aware Layout Generation with Design Prior

TL;DR

SEGA introduces a hierarchical, stepwise paradigm for content-aware layout generation that couples a coarse-level estimator with a fine-level refiner, guided by explicit design principles and feedback. The approach leverages large-language-models to perform multi-step reasoning and self-correction, improving planning fidelity on complex posters and graphics. A new GenPoster-100K dataset supports robust pre-training and cross-domain generalization, with extensive experiments showing state-of-the-art results across PKU, CGL, and Crello benchmarks. The work demonstrates that coarse-to-fine reasoning plus design-prior guidance and iterative inference significantly enhance layout quality and scalability to scenes with many elements.

Abstract

In this paper, we study the content-aware layout generation problem, which aims to automatically generate layouts that are harmonious with a given background image. Existing methods usually deal with this task with a single-step reasoning framework. The lack of a feedback-based self-correction mechanism leads to their failure rates significantly increasing when faced with complex element layout planning. To address this challenge, we introduce SEGA, a novel Stepwise Evolution Paradigm for Content-Aware Layout Generation. Inspired by the systematic mode of human thinking, SEGA employs a hierarchical reasoning framework with a coarse-to-fine strategy: first, a coarse-level module roughly estimates the layout planning results; then, another refining module performs fine-level reasoning regarding the coarse planning results. Furthermore, we incorporate layout design principles as prior knowledge into the model to enhance its layout planning ability. Besides, we present GenPoster-100K that is a new large-scale poster dataset with rich meta-information annotation. The experiments demonstrate the effectiveness of our approach by achieving the state-of-the-art results on multiple benchmark datasets. Our project page is at: https://brucew91.github.io/SEGA.github.io/

Figures (5)

• Figure 1: Distinct from traditional methods using single-stage planning model (a), we propose a hierarchical layout generation paradigm (b) guided by design prior knowledge. Accordingly, our refinement module accepts an initial layout draft from the first stage and iteratively improves it. The superiority of our method is proven by handling difficult scenes on the Crello dataset (arranging $>$ 8 elements) (c). The metric in (c) is underlay accuracy which is introduced in the experiment section.
• Figure 2: The training overview of SEGA. We introduce the training process of the coarse-level estimation model and the fine-level refinement model, as shown in (a) and (b).
• Figure 3: Qualitative visualization comparison in PKU dataset.
• Figure 4: Qualitative visualization comparison in the Crello dataset. Since this dataset is layer-independent, we present the rendered results of the layout predicted by the model.
• Figure 5: The impact of iterations of SEGA. As iteration rounds increases, the model performance gradually improves and tends towards saturation. The two metrics in the left figure are better with higher values, while it is the opposite in the right figure.