|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
\documentclass[final]{beamer} |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
\usepackage[T1]{fontenc} |
|
|
\usepackage{lmodern} |
|
|
\usepackage[size=custom,width=120,height=72,scale=1.05]{beamerposter} |
|
|
\usetheme{gemini} |
|
|
\usecolortheme{cam} |
|
|
\usepackage{graphicx} |
|
|
\usepackage{booktabs} |
|
|
\usepackage[numbers]{natbib} |
|
|
\usepackage{tikz} |
|
|
\usepackage{pgfplots} |
|
|
\pgfplotsset{compat=1.14} |
|
|
\usepackage{anyfontsize} |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
\newlength{\sepwidth} |
|
|
\newlength{\colwidth} |
|
|
\setlength{\sepwidth}{0.025\paperwidth} |
|
|
\setlength{\colwidth}{0.3\paperwidth} |
|
|
|
|
|
\newcommand{\separatorcolumn}{\begin{column}{\sepwidth}\end{column}} |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
\title{Paper2Poster: \\ Towards Multimodal Poster Automation from Scientific Papers} |
|
|
|
|
|
\author{Wei Pang, Kevin Qinghong Lin, Xiangru Jian, Xi He, Philip Torr} |
|
|
|
|
|
\institute[shortinst]{1 University of Waterloo, 2 National University of Singapore, 3 University of Oxford} |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
\footercontent{ |
|
|
\href{https://www.example.com}{https://www.example.com} \hfill |
|
|
ABC Conference 2025, New York --- XYZ-1234 \hfill |
|
|
\href{mailto:alyssa.p.hacker@example.com}{alyssa.p.hacker@example.com}} |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
\setbeamerfont{title}{size=\huge} |
|
|
\setbeamerfont{author}{size=\Large} |
|
|
\setbeamerfont{institute}{size=\large} |
|
|
\setbeamerfont{block title}{size=\Large} |
|
|
\setbeamerfont{block body}{size=\large} |
|
|
\setbeamerfont{caption}{size=\small} |
|
|
\setlength{\abovecaptionskip}{4pt} |
|
|
\setlength{\belowcaptionskip}{3pt} |
|
|
\begin{document} |
|
|
|
|
|
|
|
|
|
|
|
\addtobeamertemplate{headline}{} |
|
|
{ |
|
|
\begin{tikzpicture}[remember picture,overlay] |
|
|
\node [anchor=north west, inner sep=3cm] at ([xshift=0.0cm,yshift=1.0cm]current page.north west) |
|
|
{\includegraphics[height=4.5cm]{logos/cambridge-reversed-color-logo.eps}}; |
|
|
|
|
|
\node[anchor=north east, inner sep=2.0cm] at ([xshift=-2.0cm,yshift=0.0cm]current page.north east) |
|
|
{\includegraphics[height=6.0cm]{logo.png}}; |
|
|
\end{tikzpicture} |
|
|
} |
|
|
|
|
|
\begin{frame}[t] |
|
|
\begin{columns}[t] |
|
|
\separatorcolumn |
|
|
\begin{column}{\colwidth} |
|
|
\begin{block}{Introduction} |
|
|
Academic posters are crucial for \textbf\{scientific communication\}, allowing rapid dissemination of key findings. Unlike slide decks, posters must condense entire papers into a single page, requiring \textit\{multi-modal context handling\}, \textcolor\{red\}\{tight text-graphics interleaving\}, and \textcolor\{red\}\{spatial constraint respect\}. Existing VLM- or LLM-only approaches lack explicit visual feedback, making it difficult to maintain logical flow and legibility. |
|
|
\end{block} |
|
|
|
|
|
\begin{block}{Benchmark \& Metrics} |
|
|
We introduce the \textbf\{Paper2Poster Benchmark\}, the first benchmark for poster generation, evaluating outputs on \textcolor\{blue\}\{Visual Quality\}, \textcolor\{blue\}\{Textual Coherence\}, \textcolor\{blue\}\{Holistic Assessment\}, and \textcolor\{blue\}\{PaperQuiz\}. This benchmark pairs recent conference papers with author-designed posters, enabling systematic comparison and evaluation of generated posters. |
|
|
|
|
|
\begin{figure} |
|
|
\centering |
|
|
\includegraphics[width=0.60\linewidth]{figures/paper-picture-1.png} |
|
|
\caption{Overview of this work. We address two core challenges in scientific poster generation: Left: How to create a poster from a paper -we propose PosterAgent (Sec. 4), a framework that transforms long-context scientific papers (20K+ tokens) into structured visual posters; and Right: How to evaluate poster quality -weintroduce the Paper2Poster benchmark (Sec. 3), which enables systematic comparison between agent-generated and author-designed posters.} |
|
|
\end{figure} |
|
|
|
|
|
\end{block} |
|
|
|
|
|
\end{column} |
|
|
\separatorcolumn |
|
|
\begin{column}{\colwidth} |
|
|
\begin{block}{PosterAgent Framework} |
|
|
Our proposed \textbf\{PosterAgent\} framework is a \textit\{multi-agent pipeline\} that transforms scientific papers into structured visual posters. It consists of three components: \textcolor\{blue\}\{Parser\}, \textcolor\{blue\}\{Planner\}, and \textcolor\{blue\}\{Painter-Commenter\}. The Parser distills the paper into a structured asset library, the Planner aligns text-visual pairs into a binary-tree layout, and the Painter-Commenter loop refines each panel using VLM feedback. |
|
|
|
|
|
\begin{figure} |
|
|
\centering |
|
|
\includegraphics[width=0.78\linewidth]{figures/paper-picture-8.png} |
|
|
\caption{Illustration of the PosterAgent pipeline. Given an input paper, PosterAgent generates a structured academic poster through three modules: 1. Parser: Extracts key textual and visual assets using a combination of tools and LLM-based summarization, resulting in a structured asset library. 2. Planner: Matches assets and arranges them into coherent layouts, iteratively generating panels with a zoom-in operation. 3. Painter-Commenter: The Painter generates panel-level bullet-content along with executable code, and renders the visual output, while the Commenter-a VLM with in-context reference-provides feedback to ensure layout coherence and prevent content overflow.} |
|
|
\end{figure} |
|
|
|
|
|
\end{block} |
|
|
|
|
|
\begin{block}{Evaluation \& Results} |
|
|
Our comprehensive evaluation reveals that \textbf\{PosterAgent\} outperforms existing systems across nearly all metrics, using \textcolor\{blue\}\{87\\ |
|
|
|
|
|
\begin{figure} |
|
|
\centering |
|
|
\includegraphics[width=0.79\linewidth]{figures/paper-table-1.png} |
|
|
\end{figure} |
|
|
|
|
|
\end{block} |
|
|
|
|
|
\end{column} |
|
|
\separatorcolumn |
|
|
\begin{column}{\colwidth} |
|
|
\begin{block}{Conclusion} |
|
|
We present \textbf\{Paper2Poster\}, a new benchmark for poster generation, and the \textbf\{PosterAgent\} framework, which significantly enhances generation quality. Our findings chart clear directions for the next generation of fully automated poster-generation models, emphasizing the importance of \textit\{structured parsing\}, \textit\{hierarchical planning\}, and \textit\{visual feedback\}. |
|
|
\end{block} |
|
|
|
|
|
\end{column} |
|
|
\separatorcolumn |
|
|
\end{columns} |
|
|
\end{frame} |
|
|
|
|
|
\end{document} |
|
|
|
