RSHallu studies hallucinations in remote-sensing MLLMs with a new taxonomy, benchmark, and dual-mode checker. Provides datasets for mitigation via training and plug-and-play strategies. Improves hallucination-free rates by up to 21% on RS tasks.
Key Points
- 1.RS-specific hallucination taxonomy and eval
- 2.2k QA benchmark, 45k datasets
- 3.Domain-tailored mitigation strategies
Impact Analysis
Enhances reliability of RS MLLMs for high-stakes uses like emergency management. Maintains downstream performance.
Technical Details
Image-level inconsistencies; logit correction and prompting. Supports cloud/local checking.