Seeing Clearly without Training: Mitigating Hallucinations in Multimodal LLMs for Remote Sensing

TL;DR

This paper introduces RSHBench, a benchmark for diagnosing hallucinations in multimodal LLMs for remote sensing, and proposes RADAR, a training-free method to reduce hallucinations during inference, improving accuracy and reliability.

Contribution

The paper presents RSHBench for systematic diagnosis and RADAR for inference-time hallucination mitigation in multimodal LLMs for remote sensing.

Findings

RADAR reduces hallucinations in MLLMs during remote sensing tasks.

RADAR improves accuracy in remote sensing visual question-answering.

Extensive experiments validate RADAR's effectiveness across diverse models.

Abstract

Multimodal large language models (MLLMs) suffer from pronounced hallucinations in remote sensing visual question-answering (RS-VQA), primarily caused by visual grounding failures in large-scale scenes or misinterpretation of fine-grained small targets. To systematically analyze these issues, we introduce RSHBench, a protocol-based benchmark for fine-grained diagnosis of factual and logical hallucinations. To mitigate grounding-induced factual hallucinations, we further propose Relative Attention-Driven Actively Reasoning (RADAR), a training-free inference method that leverages intrinsic attention in MLLMs to guide progressive localization and fine-grained local reasoning at test time. Extensive experiments across diverse MLLMs demonstrate that RADAR consistently improves RS-VQA performance and reduces both factual and logical hallucinations. Code and data will be publicly available at: https://github.com/MiliLab/RADAR