Hugging Face Releases RAP: A Dataset and Benchmark for Academic Poster Generation
Summary:Hugging Face has released RAP (Reconstruct A Poster), a dataset and benchmark for generating academic posters. RAP includes over 25,000 verified paper-poster pairs from various machine learning conferences between 2022 and 2026, annotated with a three-level content taxonomy. The dataset features a fully automated derendering pipeline that converts posters into editable HTML/CSS templates, providing a new standard for training and evaluating AI-generated academic posters. RAP introduces PosterRAG
RAP: A New Benchmark for Academic Poster Generation
1. Background and Challenges
Generating academic posters is a complex visual coding task that requires compressing an entire paper onto a single page while balancing summarization, visual design, and code implementation. However, existing benchmarks are limited, primarily annotated with layout boxes, and lack objective render-and-compare evaluation. Moreover, current methods typically design layouts from scratch, making it impossible to leverage human-crafted posters as a retrievable design library or verified supervision.
2. RAP Dataset and Benchmark
RAP (Reconstruct A Poster) addresses these challenges by providing a dataset of over 25,000 verified paper-poster pairs from various machine learning conferences between 2022 and 2026, annotated with a three-level content taxonomy. It also includes a held-out benchmark of 150 test papers, each paired with its human-made poster.
3. Fully Automated Derendering Pipeline
RAP introduces a fully automated derendering pipeline that converts each poster into editable HTML/CSS templates with typed, refillable slots. This ensures that every sample is a verifiable paper-poster-markup triplet, verified by rendering against the source poster.
4. Training and Inference Methods
RAP supports two training methods:
- PosterRAG: A training-free multi-agent pipeline that retrieves an exemplar and refills its slots without altering the layout or styling.
- SFT+RL Closed: A method based on Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) for training a dedicated model.
5. Technical Innovations and Advantages
- Data Scale and Diversity: RAP's dataset is large and diverse, providing a rich set of training samples from various conferences.
- Automated Derendering: The fully automated derendering pipeline ensures data consistency and verifiability.
- Multi-Method Support: Offers both retrieval-based and model-based generation methods to cater to different needs.
6. Developer Recommendations
RAP provides a powerful toolkit for AI developers. It is recommended to use PosterRAG for rapid prototyping and SFT+RL Closed for fine-grained model training. Additionally, developers should pay attention to the details of the derendering process to ensure the consistency of the generated posters in both visual and content aspects.
Conclusion
The release of RAP provides a new benchmark and data support for the academic poster generation field, driving the application and development of AI in visual coding tasks.
— END —Source: Hugging Face Daily Papers (2026-10-16)
Tags: #Hugging Face #Dataset #Academic Poster Generation #Visual Coding #AI Generation
Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.
Community Comments