Image-aware Layout Generation with User Constraints for Poster Design

TL;DR

This paper presents a deep learning approach for automatic, user-constrained poster layout generation that incorporates image-awareness and partial layout constraints, achieving state-of-the-art results.

Contribution

It introduces a novel model with attribute and partial layout constraints, including new loss functions and a diversification technique for flexible poster design.

Findings

The model effectively generates layouts satisfying user constraints.

It outperforms existing methods in quantitative and qualitative evaluations.

The approach enables diverse, image-aware poster layouts.

Abstract

Graphic layout is essential in poster generation. Professionals often need to design different layouts for a product image, to ensure they meet specific user requirements. This paper focuses on utilizing a deep-learning model to automatically generate image-aware layouts with user-defined constraints, including layout attributes and partial layouts. Layout attribute constraints require generated layouts to include and exclude elements of specified classes, such as text, logos, underlays, and embellishments. Our model represents different attributes by sampling multidimensional Gaussian noise with different means, and we propose an attribute-consistent loss and an attribute-disentangled loss to ensure that the generated layout satisfies the specified attribute. Partial layout constraints provide our model with incomplete layout information to guide the generation of the remaining elements. We design a partial-constraint loss to incorporate the provided partial layout. Furthermore, we introduce a random mask to diversify the partial layout constraints, which can encourage the model to learn more general latent representations of the provided partial layouts. Both quantitative and qualitative evaluations demonstrate that our model can generate different image-aware layouts according to various user constraints while achieving state-of-the-art performance.