Reproduce “AutoPoster: A Highly Automatic and Content-aware Design System for Advertising Poster Generation” 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
Jinpeng Lin, Min Zhou, Ye Ma, Yifan Gao, Chenxi Fei, Yangjian Chen, Zhang Yu, Tiezheng Ge
AutoPoster is a highly automatic and content-aware system developed by Alibaba Group for generating advertising posters from product images and titles. It achieved an 8.24% increase in Click-Through Rate and an 8.25% increase in Revenue Per Mille in an online A/B test, alongside constructing the first large-scale poster generation dataset with visual attribute annotations.
Key insights
- Jointly predicting multiple correlated style attributes (colors, font) through multi-task learning significantly improves overall design harmony and accuracy.
- Non-autoregressive decoding for style attribute prediction is more effective than autoregressive methods, aligning with the holistic nature of design.
- A robust image cleaning and retargeting pipeline is crucial for handling diverse real-world input images and adapting them to various poster sizes without visual distortion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.