DroneIQA-VLE: Multi-Task Drone Image Quality Assessment via Vision-Language Ensemble
2026-07-01 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors developed DroneIQA-VLE, a system to judge the quality of drone images, specifically focusing on the main subject and background. They combined two different methods: one using specialized vision models and another using a large language model adapted for image quality rating. By averaging results from both, their system provides an overall quality score. Their approach ranked second in a 2026 challenge, showing it works well for this task.
Image Quality AssessmentDrone ImagesMulti-task RegressionVision EncodersLarge Language ModelsLoRA (Low-Rank Adaptation)Multimodal ModelsEnsembling MethodsUAV (Unmanned Aerial Vehicle)
Authors
Wei Sun, Weixia Zhang, Hongjian Zhan, Mingkai Lu, Yixuan Gao, Guangtao Zhai
Abstract
We present DroneIQA-VLE, our solution to the ICME 2026 Drone-IQA Grand Challenge on Target-aware Image Quality Assessment for Low-altitude UAV Images. The framework jointly predicts global, target, and background quality scores by ensembling two complementary pipelines: (1) SigLIP2 vision encoders with multi-task regression heads, and (2) a LoRA-adapted Qwen3.5-9B multimodal large language model for quality score regression. The final global quality prediction is obtained by arithmetically averaging the outputs of both pipelines. Our method achieves 2nd place in the challenge, demonstrating its effectiveness. The code is available at https://github.com/sunwei925/DroneIQA-VLE.