Researchers propose R^3, a novel framework designed to rectify textual violations in video advertisements while preserving the advertiser's original semantic intent. The system addresses the challenge of manual rectification at scale by integrating text recognition, rewriting, and re-rendering for industrial deployment.

  • An experience-driven data synthesis framework bootstraps high-quality supervision via a group-relative compliance experience extractor.
  • A curriculum reinforcement learning strategy with hierarchical rewards enforces compliance while maximizing semantic consistency.
  • Comprehensive video rectification capabilities seamlessly integrate text recognition, rewriting, and re-rendering.

Extensive experiments on industrial datasets and online A/B testing demonstrate that R^3 significantly outperforms state-of-the-art baselines, achieving an optimal trade-off between violation rectification and intent preservation.