Benchmark RadarFlameVQA
FlameVQA is a multiple-choice visual question answering benchmark for UAV-based wildfire monitoring, built on FLAME 3 with paired RGB and radiometric thermal images. It includes 34 questions per image across six capability groups, covering detection, localization, coverage estimation, cross-modal reasoning, and flight planning. Labels are generated via MLLM assistance, deterministic thermal rules, and human auditing. The dataset and code are open-source.
- Released
- 2026-06-25
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
FlameVQA addresses the lack of benchmarks for evaluating vision-language models in safety-critical wildfire scenarios where RGB-only interpretation is insufficient. It provides a standardized evaluation for capabilities like smoke detection and coverage estimation, which are critical for practical deployment of MLLMs in disaster monitoring.
Motivation
Wildfire monitoring from UAVs requires reliable reasoning over complex aerial scenes, where smoke, scale variation, and occlusions often limit RGB-only interpretation.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.