Reproduce “AutoPP: Towards Automated Product Poster Generation and Optimization” 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
Jiahao Fan, Yuxin Qin, Wei Feng, Yanyin Chen, Yaoyu Li, Ao Ma, Yixiu Li, Li Zhuang and 5 more
JD.COM researchers developed AutoPP, an end-to-end system for automated product poster generation and online optimization using a unified design module and element-isolated preference learning. The system achieves a 4.49% relative Click-Through Rate (CTR) improvement in real-world scenarios while producing visually superior and text-accurate posters compared to previous methods.
Key insights
- A unified design approach, jointly reasoning about background, text, and layout, yields more coherent and higher-quality product posters than fragmented, multi-stage generation pipelines.
- Element-isolated optimization (IDPO) effectively attributes online performance gains to specific design components, leading to more targeted, interpretable, and efficient CTR improvements than holistic optimization methods.
- Robust and efficient visual text rendering, crucial for marketing content, can be achieved by integrating glyph image token encoding and an OCR perceptual loss within a diffusion model's training objective.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.