PosterForest: Hierarchical Multi-Agent Collaboration for Scientific Poster Generation (2508.21720v1)
Abstract: We present a novel training-free framework, \textit{PosterForest}, for automated scientific poster generation. Unlike prior approaches, which largely neglect the hierarchical structure of scientific documents and the semantic integration of textual and visual elements, our method addresses both challenges directly. We introduce the \textit{Poster Tree}, a hierarchical intermediate representation that jointly encodes document structure and visual-textual relationships at multiple levels. Our framework employs a multi-agent collaboration strategy, where agents specializing in content summarization and layout planning iteratively coordinate and provide mutual feedback. This approach enables the joint optimization of logical consistency, content fidelity, and visual coherence. Extensive experiments on multiple academic domains show that our method outperforms existing baselines in both qualitative and quantitative evaluations. The resulting posters achieve quality closest to expert-designed ground truth and deliver superior information preservation, structural clarity, and user preference.
Summary
- The paper presents a novel framework, PosterForest, that leverages hierarchical document structures through a Poster Tree and multi-agent collaboration for automated poster generation.
- The methodology transforms papers into a structured Poster Tree using Raw, Content, and Layout phases, enabling effective integration of textual and visual elements.
- The framework outperforms existing SPG approaches in content fidelity and layout precision, as demonstrated by benchmark comparisons and ablation studies.
PosterForest: Hierarchical Multi-Agent Collaboration for Scientific Poster Generation
Introduction
The paper "PosterForest: Hierarchical Multi-Agent Collaboration for Scientific Poster Generation" introduces a non-training framework for automating the generation of scientific posters. Unlike prior approaches that often ignore hierarchical document structures and the integration of visual and textual elements, PosterForest explicitly models these features. The method utilizes a Poster Tree, a hierarchical intermediate representation that captures document structure and visual-textual relationships, alongside a multi-agent strategy for content summarization and layout planning. This approach jointly optimizes logical consistency, content fidelity, and visual coherence.
Background
Existing automated scientific poster generation (SPG) techniques, such as PGM, NCE, and PostDoc, rely on heuristic text and figure extraction methods. More advanced methods like P2P and Paper2Poster employ multi-agent systems requiring explicit model training. However, these methods have limitations, including a lack of understanding of the document's hierarchical structure and a weak integration of content and layout. PosterForest addresses these gaps by emphasizing the hierarchical organization and multi-agent collaboration without pre-trained models.
Methodology
Hierarchical Paper-to-Poster
PosterForest transforms a paper into a Poster Tree. Initially, a Raw Document Tree is created, followed by a Content Tree, which selects and summarizes essential information. The Layout Tree is then initialized to define spatial arrangements, resulting in the Poster Tree. This structure encapsulates both content and layout details, addressing the limitations of earlier methods that treat text and visuals as separate entities.
Figure 1: Overview of PosterForest.
Collaborative Poster Tree Optimization
The Poster Tree undergoes iterative refinement through collaboration between a Content Agent and a Layout Agent. Each agent specializes in tasks like content summarization or visual material placement. The agents iteratively update the Tree, improving logical consistency and integration of textual and visual information. Through node-level analysis and iterative tree refinement, the agents align content and layout effectively.
Figure 2: Modification Planning. The Poster Tree and layout are iteratively updated through the shared decision of layout and Content Agent.
Implementation and Results
PosterForest is benchmarked against competing models like P2P and Paper2Poster using paper-poster pairs from diverse academic fields. The framework demonstrates superior performance in information preservation and layout precision. Ablation studies highlight the effectiveness of hierarchical structuring and agent collaboration in achieving these results.
Figure 3: Qualitative Comparison. Posters generated with the GPT-4o framework of baseline methods and PosterForest, based on papers spanning different AI fields (computer vision, NLP, RL), along with the original posters (GT) created by the authors.
Discussion and Future Work
While PosterForest advances the state of SPG, challenges remain. One limitation is the framework's performance in handling papers with complex figures, where current parsing may lead to errors. Future work should enhance parsing capabilities and develop more robust automated metrics for quality assessment. Additionally, expanding the framework's ability to evaluate the semantic importance and spatial consideration of figures could further improve poster generation quality.
Conclusion
PosterForest presents an innovative approach to SPG by embracing document hierarchy and fostering multi-agent collaboration. The resulting posters exhibit enhanced logical consistency, content fidelity, and visual coherence. By addressing significant limitations of preceding methods, PosterForest establishes a new benchmark in the automated generation of scientific posters. Further advancements in parsing techniques and evaluation metrics are anticipated to refine the efficacy of this framework.
Collections
Sign up for free to add this paper to one or more collections.
Paper Prompts
Sign up for free to create and run prompts on this paper using GPT-5.