Design-o-meter: Towards Evaluating and Refining Graphic Designs

TL;DR

This work introduces Design-o-meter, a data-driven methodology to quantify the goodness of graphic designs, and is the first approach that scores and refines designs in a unified framework despite the inherent subjectivity and ambiguity of the setting.

Abstract

Graphic designs are an effective medium for visual communication. They range from greeting cards to corporate flyers and beyond. Off-late, machine learning techniques are able to generate such designs, which accelerates the rate of content production. An automated way of evaluating their quality becomes critical. Towards this end, we introduce Design-o-meter, a data-driven methodology to quantify the goodness of graphic designs. Further, our approach can suggest modifications to these designs to improve its visual appeal. To the best of our knowledge, Design-o-meter is the first approach that scores and refines designs in a unified framework despite the inherent subjectivity and ambiguity of the setting. Our exhaustive quantitative and qualitative analysis of our approach against baselines adapted for the task (including recent Multimodal LLM-based approaches) brings out the efficacy of our methodology. We hope our work will usher more interest in this important and pragmatic problem setting.

Figures (9)

• Figure 1: The figure presents an overview of Design-o-meter. It takes a design as input. The scorer evaluates the design and provides a unified design score. The refiner refines the design with the help of the design score to improve its aesthetic appeal.
• Figure 2: A visual illustration of SWAN: Design Specific Crossover with Smart Snapping. Given two parents, SWAN first randomly decides which element to pick from either of the parents to generate the child. Then, it copies over the content from the first parent to the current canvas. Next, it identifies potential area within the canvas to host elements from the second parent, guided by grid-lines. Finally, the content from the second parent is 'snapped' into the identified areas by changing its attributes. This allows SWAN to generate better results when compared with regular crossover.
• Figure 3: Samples from Crello dataset and their layout encodings.
• Figure 4: Qualitative results of Design-o-meter refining graphic designs. The top sub-figure refines all elements of a design (Refine-All setting), while the bottom one refines a single text box (Refine-Text setting). We see that our approach is able to automatically improve the position and scale parameters of design elements, making them more visually appealing.
• Figure 5: These occlusion-based sensitivity maps show the locations of the design with positive (red) or negative (blue) impacts on a design score prediction. The numbers correspond to scores.