ZhanYang-nwpu/UAVIT-1M

🎉 UAVIT-1M stands as the first largest and most comprehensive instruction tuning dataset to enhance low-altitude UAV visual understanding abilities, supporting 11 image-level and region-level tasks.

<div align="center"> <img src="https://github.com/ZhanYang-nwpu/SkyEyeGPT/blob/main/images/UAVIT-1M.png?raw=true" width="20%"/> </div>

📢 News

This is an ongoing project. We will be working on improving it.

• —📦 Complete evaluation code coming soon! 🚀
• —📦 Detailed low-altitude UAV MLLMs model inference tutorial coming soon! 🚀
• —[05/13/2025] 🚀 We release our UAVBench benchmark and UAVIT-1M instruct tunning data to huggingface. UAVBench
• —[05/13/2025] 🚀 We release 3 low-altitude UAV Multi-modal Large Language Model baselines. 🤖LLaVA1.5-UAV, 🤖MiniGPTv2-UAV, 🤖GeoChat-UAV

🎆 UAVIT-1M Instruct Tuning Dataset

UAVIT-1M consists of approximately 1.24 million diverse instructions, covering 789k multi-scene low-altitude UAV images and about 2,000 types of spatial resolutions with 11 distinct tasks.

💎 UAVBench Benchmark

To evaluate existing MLLMs’ abilities in low-altitude UAV vision-language tasks, we introduce the UAVBench Benchmark. The UAVBench comprises about 966k high-quality data samples and 43 test units, across 10 tasks at the image-level and region-level, covering 261k multi-spatial resolution and multi-scene images. UAVBench and UAVIT-1M feature pure real-world visual images and rich weather conditions, and involve manual sampling verification to ensure high quality.

For more information, please refer to our 🏠Homepage.

📨 Contact

If you have any questions about this project, please feel free to contact zhanyangnwpu@gmail.com.