RAP releases dataset for converting papers to posters using declarative markup
benchmarks
| Source: HF Papers | Original article
Researchers unveil RAP, a new dataset enabling paper-to-poster generation via declarative markup drawn from human‑crafted exemplars.
A new dataset and benchmark called **RAP** has been released to spur research on automatically turning academic papers into conference posters. The authors present more than 25 000 verified paper‑poster pairs drawn from machine‑learning conferences spanning 2022‑2026, each annotated with a three‑level content taxonomy. A held‑out test set of 150 papers, paired with the posters created by the original authors, forms the core of the benchmark.
The task is framed as “visual coding in miniature”: a system must compress an entire research article onto a single page, simultaneously handling summarisation, visual layout and the low‑level code that drives the poster’s design. Existing benchmarks have largely ignored this blend of natural‑language understanding and graphic generation, leaving a gap for models that can bridge the two. By providing a large, curated corpus and a clear evaluation protocol, RAP gives researchers a concrete target for developing multimodal agents that can both understand scientific content and produce polished visual artefacts.
The release matters because automated poster creation could streamline the dissemination of research, reduce the time scientists spend on design work, and open up new avenues for AI‑driven scientific communication tools. Moreover, the dataset’s taxonomy may serve as a scaffold for future work on hierarchical content planning and layout optimisation in multimodal generation.
Watch for early results from teams applying large language models, diffusion‑based image generators and hybrid visual‑coding frameworks to the RAP benchmark. Competitions or leaderboards are likely to emerge, and integration with existing visual‑coding platforms could turn the dataset into a standard testbed for next‑generation AI assistants that blend text summarisation with graphic design.
Sources
Back to AIPULSEN