Search papers, labs, and topics across Lattice.
This study addresses the challenges of aerial image understanding and reasoning using Multimodal Large Language Models (MLLMs) by creating UAVQA-Bench, a comprehensive benchmark featuring 1,500 human-annotated QA pairs across various UAV datasets. The evaluation of multiple MLLMs reveals critical failure modes such as domain-toolset mismatch and unchecked error propagation, leading to the development of UAV-MAS, a training-free multi-agent system designed to enhance performance through specialized modules for perception and reasoning. The proposed system achieves an overall accuracy of 77.0% on UAVQA-Bench, outperforming existing models, including Gemini 3 Pro, by a notable margin.
A training-free multi-agent system for UAV image understanding not only surpasses leading models in accuracy but also addresses fundamental reasoning failures in MLLM applications.
Multimodal Large Language Model (MLLM)-based UAV aerial image understanding and reasoning is essential for aerial intelligence yet poses distinct challenges arising from extreme scale variation, arbitrary camera orientations, and high object density. Despite growing interest, existing evaluations remain fragmented across individual datasets and narrow tasks, leaving a critical gap in unified assessment of UAV understanding and reasoning capabilities. To fill this gap, we construct UAVQA-Bench, a benchmark of 1,500 human-annotated QA pairs drawn from 13 public UAV datasets, covering 6 capability dimensions and 16 tasks in both multiple-choice and visual grounding formats. Systematic evaluation of a broad range of open-source and closed-source MLLMs as well as agent-based systems on UAVQA-Bench identifies three key failure modes: domain-toolset mismatch, unchecked error propagation, and static reasoning. Motivated by these findings, we propose UAV-MAS, a training-free multi-agent system for MLLM-based UAV aerial image understanding and reasoning, comprising a Domain-Specific Perception Engine (DSPE) that routes queries to task-appropriate visual tools, a Context-Aware Iterative Refinement module (CAIR) that validates intermediate reasoning to curb error accumulation, and a Difficulty-Aware Adaptive Search mechanism (DAAS) that adjusts search depth to question difficulty. UAV-MAS with a 32B open-source MLLM achieves 77.0% overall accuracy on UAVQA-Bench, surpassing Gemini 3 Pro by 4.0\%, while the 8B variant improves 8.7\% over its base model.
