A novel framework for logical image fraud detection using cross-sample reasoning and LLMs.
- 🔍 Cross-sample reasoning over document images
- ✅ LLM-powered factual rule generation and verification
- 🗂️ Released CrossCred: a real-world dataset covering diverse types of fraud images in the financial sector, with detailed human annotations to support evaluation of cross-sample image fraud detection methods
- 🚧 The CrossCred benchmark is under de-identification and will be released soon.
- 🌐 2025-05-25: Project homepage launched.
- 🎉 2025-05-15: CSIAD accepted at ACL 2025!