Reproduce “SciPostGen: Bridging the Gap between Scientific Papers and Poster Layouts” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shun Inadumi, Shohei Tanaka, Tosho Hirasawa, Atsushi Hashimoto, Koichiro Yoshino, Yoshitaka Ushiku
A new large-scale dataset, SciPostGen, containing 18,097 paired scientific papers and their corresponding poster layouts, facilitates data-driven understanding and generation of scientific poster designs. It underpins a retrieval-augmented framework that uses an LLM to generate poster layouts consistent with paper structures and adaptable to user constraints, demonstrating improved layout quality and element consistency.
Key insights
- Analysis of the SciPostGen dataset reveals moderate correlations between paper characteristics (e.g., text length, number of figures/tables) and poster layout element counts and areas.
- Retrieved high-quality poster layouts, serving as prompts for an LLM generator, consistently improve layout consistency and element distribution compared to random examples.
- The LLM generator effectively refines retrieved layouts and integrates user-specified partial constraints, demonstrating its complementary role in layout customization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.