Latest News
About the Challenge
The rapid advancement of AIGC has fundamentally lowered the barrier to digital content manipulation, shifting document tampering from "visible visual modifications" to a sophisticated fusion of visual forgery and semantic manipulation. In the AGI era, covert techniques—such as visually similar character substitution—not only evade traditional detection but also pose cascading poisoning risks across OCR–LLM pipelines. As global AI regulations tighten, simple binary authenticity judgments are no longer sufficient; there is an urgent need for explainable, reasoning-based forensic capabilities.
To address these critical security threats, we launch GenText-Forensics, the first AI security challenge dedicated to global multilingual text-image forensics. The challenge reframes document forgery detection as a unified generative "forgery analysis" task, aiming to bridge the gap in evidence traceability and reasoning insight by producing structured forensic reports.
Supported by RealText-V2, the world's first large-scale multilingual document forgery benchmark, this challenge provides a solid data foundation featuring:
- 20K+ Multimodal Samples: Spanning 6 languages and 6 major domains including finance, healthcare, and education.
- 100+ Attack Methods: Covering granularity levels from character substitution to sentence-level semantic manipulation.
- Expert-Level Annotations: Integrating pixel-level localization, attack type classification, and natural language explanations.
Forgery Analysis Report Generation
- Requiring the generation of comprehensive reports that integrate detection, spatial grounding, and natural language explanation.