This article explains Alibaba's front‑end intelligent system that automatically generates UI code by extracting design metadata, recognizing basic components with a YOLO‑based model, and refining predictions, detailing the pipeline from sample creation to model evaluation and future enhancements.
This article explains how Alibaba's Frontend Intelligent project uses design‑to‑code techniques, component recognition with YOLO, and a full pipeline of sample creation, model training, evaluation, and prediction refinement to automatically generate a large portion of the Double‑11 event code.
The paper presents a structured layout‑information framework combined with a guided‑line “leader‑follower” algorithm that represents UI controls as Connection objects and matches them via attribute and vector similarity, enabling fast identification of recurring business components and duplicate GridView items without extensive retraining, thereby enhancing code reuse in UI2CODE projects.
