What is Paper2Poster
Paper2Poster is an innovative academic framework launched by the University of Waterloo in Canada, the National University of Singapore, and other institutions, which generates posters from scientific papers based on multimodal automation technology. Paper2Poster introduces PosterAgent, a top-down multi-agent system that supports compressing and transforming long paper content into structured visual posters. The system achieves efficient generation based on a Parser, a Planner, and a Painter–Commenter loop. Paper2Poster introduces the PaperQuiz evaluation method, which simulates readers answering questions to measure the poster's ability to convey its core content. Paper2Poster performs excellently in visual quality and textual coherence, can significantly improve generation efficiency, and provides an efficient, low-cost solution for academic poster creation.
Main Features of Paper2Poster
- Long-text compression: Compresses multi-page scientific papers into a single-page poster, preserving the paper's core content.
- Multimodal content processing: Extracts multimodal content such as text, charts, and images from the paper and reasonably integrates them into the poster.
- Layout optimization: Generates a visually appealing and logically clear poster layout, ensuring content is reasonably distributed within limited space.
- Visual quality improvement: Optimizes the poster's visual effects based on a visual feedback mechanism, improving the poster's readability and appeal.
- Evaluation and optimization: Evaluates the poster's ability to convey its core content based on PaperQuiz and optimizes it according to feedback.
Technical Principles of Paper2Poster
- Parser: Uses tools (such as MARKER and DOCLING) to convert PDFs into Markdown format, and generates a structured JSON-format asset library based on an LLM.
- Planner: Aligns the text and visual elements in the asset library generated by the parser to generate a binary tree layout. Based on a binary tree layout strategy, it estimates the size of each panel according to content length, maintaining reading order and spatial balance. It uses an LLM for semantic matching to align each visual element with the most relevant text section.
- Painter–Commenter: Generates the specific content of each panel and optimizes the panel layout based on a visual feedback mechanism. The painter aligns text and images and generates executable code, rendering panels with the python-pptx library. The commenter is a VLM that provides feedback based on zoom-in reference prompts, ensuring that panel content does not overflow and that the layout is reasonable.
Paper2Poster Project Address
- Project official website：https://paper2poster.github.io/
- GitHub repository：https://github.com/Paper2Poster/Paper2Poster
- HuggingFace model library：https://huggingface.co/datasets/Paper2Poster/Paper2Poster
- arXiv technical paper：https://arxiv.org/pdf/2505.21497
Application Scenarios of Paper2Poster
- Academic conferences: Helps researchers quickly transform papers into posters for conference presentations, saving time and effort.
- Academic reports: The generated posters serve as auxiliary materials for academic reports, helping the audience better understand the research content.
- Scientific research achievement display: In scientific research institutions or laboratories, displays the latest research results to facilitate peer exchange and learning.
- Education field: Teachers generate posters for teaching use, helping students understand complex academic concepts more intuitively.
- Online academic platforms:Provide automated poster generation tools for online academic platforms, enhance user experience, and promote academic exchange.