Where Does It Exist from the Low-Altitude: Spatial Aerial Video Grounding

🎉 Accepted by Neural Information Processing Systems (NeurIPS 2025) 🎉

Please share a STAR ⭐ if this project does help

You can focus on multimodal large language model (Vision-Language) for UAV here

📢 Latest Updates

• 📦 Paper and codebase are coming soon! 🚀
• Detailed codebase and model tutorial coming soon!
• Dec-09-2025: Our UAV-SVG dataset is released. 🔥🔥
• Sep-19-2025: SAVG paper is accepted by NeurIPS 2025. 🔥🔥

---

🌋 UAV-SVG Dataset

Dataset Preparation Tutorial:

• Step1: Please download the JSON annotation file: https://huggingface.co/datasets/ZhanYang-nwpu/UAV-SVG .
• Step2: Please download the aerial video from the official link of WebUAV-3M: https://github.com/983632847/WebUAV-3M?tab=readme-ov-file#dataset-download. [Note: the data size: 819G]

Note that the dataset is ONLY permitted to be used for research.

📜Reference

If you found this code useful, please cite the paper. Welcome :+1:Fork and Star:+1:, then I will let you know when we update.

@inproceedings{zhandoes,
  author={Zhan, Yang and Yuan, Yuan},
  title={Where Does It Exist from the Low-Altitude: Spatial Aerial Video Grounding},
  booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems},
  year={2025}
  }

🙏Acknowledgments

Our UAV-SVG is constructed based on the WebUAV-3M aerial video object tracking dataset. We thank to the authors for releasing the dataset. I would like to thank Xiong zhitong and Yuan yuan for helping the manuscript.

License

Licensed under a Creative Commons Attribution-NonCommercial 4.0 International for Non-commercial use only. Any commercial use should get formal permission first.