From Transient Prompts to Persistent Control: Scientific Poster Generation via Recursive Semantic-Geometric Contracts

Problem

Existing methods for scientific poster generation rely on transient prompts, which can lead to inconsistencies and drift in requirements across content and layout modules. This paper addresses this gap by proposing a more robust framework that maintains persistent control over the generation process. The work is presented as a preprint and has not undergone peer review.

Method

The authors introduce PosterVisor, a novel framework designed to improve the generation of scientific posters. The core components of PosterVisor include:

• Orchestrator: This component compiles various rubrics into a Semantic-Geometric Contract (SGC), which serves as a structured guideline for the poster generation process.
• Recursive Contract Enforcement (RCE): This mechanism triggers checks at different stages of the generation process to ensure adherence to the SGC, thereby maintaining consistency and quality.
• The implementation supports output in HTML/CSS format and generates editable PPTX files, allowing for flexibility in the final presentation of the posters.

Results

The performance of PosterVisor is evaluated against existing methods:

• PosterVisor-PPT QA accuracy: 64.47%, compared to PosterGen which achieved 58.53%.
• In terms of human preference, 72.5% of non-tied comparisons favored PosterVisor, with no baseline reported for this metric.
• A secondary study indicated that PosterVisor yields higher means in both VLM Overall and PaperQuiz metrics, although specific baseline comparisons are not provided.

Limitations

The authors do not report any limitations in their work. However, the absence of a comprehensive baseline for human preference and secondary study metrics may limit the interpretability of the results.

Why it matters

The development of PosterVisor has significant implications for the field of automated content generation, particularly in scientific communication. By addressing the limitations of transient prompts, this framework enhances the reliability and quality of generated posters, which can facilitate better dissemination of scientific knowledge. The recursive enforcement of semantic-geometric contracts could also inspire future research into more structured approaches for other forms of content generation.