Vane3RS/DisasterM3
DisasterM3: A Remote Sensing Vision-Language Dataset for Disaster Damage Assessment and Response Junjue Wang*, Weihao Xuan*, Heli Qi, Zhihao Liu, Kunyi Liu, Yuhan Wu, Hongruixuan Chen, Jian Song Junshi Xia, Zhuo Zheng, Naoto Yokoya† * Equal Contributions † Corresponding Author Paper: https://arxiv.org/abs/2505.21089 Code: https://github.com/Junjue-Wang/DisasterM3 Highlights DisasterM3 includes 26,988 bi-temporal satellite images and 123k instruction pairs across… See the full description on the dataset page: https://huggingface.co/datasets/Vane3RS/DisasterM3.
119
1---2license: cc-by-nc-sa-4.03task_categories:4- visual-question-answering5- image-segmentation6- image-to-text7- image-classification8language:9- en10tags:11- vision-language12- remote-sensing13- disaster14- multi-task15pretty_name: >-16 A multi-hazard, multi-sensor, and multi-task vision-language dataset for17 global-scale disaster assessment and response.18size_categories:19- 1B<n<10B20---21 22 23<h2 align="center">24 DisasterM3: A Remote Sensing Vision-Language Dataset for Disaster Damage Assessment and Response25</h2>26 27<h5 align="center"><a href="https://junjue-wang.github.io/homepage/">Junjue Wang*</a>,28<a href="https://weihaoxuan.com">Weihao Xuan*</a>,29Heli Qi, Zhihao Liu, Kunyi Liu, Yuhan Wu, <a href="https://chrx97.com/"> Hongruixuan Chen</a>,30<a href="https://jtrneo.github.io/"> Jian Song</a></h5>31<h5 align="center">32Junshi Xia, <a href="https://zhuozheng.top/">Zhuo Zheng</a>, <a href="https://naotoyokoya.com/">Naoto Yokoya†</a></h5>33 34<h5 align="center">35* Equal Contributions36† Corresponding Author</h5>37 38`Paper`: https://arxiv.org/abs/2505.2108939 40`Code`: https://github.com/Junjue-Wang/DisasterM341 42 43<div align="center">44 <img src="https://github.com/Junjue-Wang/resources/blob/main/DisasterM3/task_taxonomy.png?raw=true">45</div>46 47## Highlights48DisasterM3 includes 26,988 bi-temporal satellite images and 123k instruction pairs across 5 continents, with three characteristics:491. Multi-hazard: 36 historical disaster events with significant impacts, which are categorized into 10 common natural and man-made disasters502. Multi-sensor: Extreme weather during disasters often hinders optical sensor imaging, making it necessary to combine Synthetic Aperture Radar (SAR) imagery for post-disaster scenes513. Multi-task: 9 disaster-related visual perception and reasoning tasks, harnessing the full potential of VLM's reasoning ability52 53 54## News55- 2025/10/23, We released the DisasterM3 instruct set.56- 2025/10/17, We released the benchmark set of DisasterM3.57- 2025/09/22, We are preparing the dataset and code.58- 2025/09/22, Our paper got accepted by NeurIPS 2025.59 60 61## Benchmark62 63Please run this code for benchmarking the DisasterM3 dataset.64Two examples:65Qwen2.5 VL:66```67python disaster_m3/pyscripts/run_vllm.py --model_id Qwen/Qwen2.5-VL-7B-Instruct --subset bearing_body68```69InternVL3:70```71python disaster_m3/pyscripts/run_vllm.py --model_id OpenGVLab/InternVL3-78B --subset report72```73 74 75## Citation76If you use DisasterM3 in your research, please cite our following papers.77```text78 @article{wang2025disasterm3,79 title={DisasterM3: A Remote Sensing Vision-Language Dataset for Disaster Damage Assessment and Response},80 author={Wang, Junjue and Xuan, Weihao and Qi, Heli and Liu, Zhihao and Liu, Kunyi and Wu, Yuhan and Chen, Hongruixuan and Song, Jian and Xia, Junshi and Zheng, Zhuo and Yokoya, Naoto},81 booktitle={Proceedings of the Neural Information Processing Systems},82 year={2025}83}84```85 86## Acknowledgments87This dataset builds upon the following excellent open datasets:88- **xBD dataset** by Ritwik Gupta89 - [Paper](https://openaccess.thecvf.com/content_CVPRW_2019/html/cv4gc/Gupta_Creating_xBD_A_Dataset_for_Assessing_Building_Damage_from_Satellite_CVPRW_2019_paper.html)90 - [Dataset](https://xview2.org/dataset)91 - License: [CC BY-NC-SA 4.0]92 93- **BRIGHT dataset** by Hongruixuan Chen94 - [Repository](https://github.com/ChenHongruixuan/BRIGHT)95 - License: [CC BY-NC 4.0]96 97 98## License99All images and their associated annotations in DisasterM3 can be used for academic purposes only,100<font color="red"><b> but any commercial use is prohibited.</b></font>101 102<a rel="license" href="https://creativecommons.org/licenses/by-nc-sa/4.0/deed.en">103<img alt="知识共享许可协议" style="border-width:0" src="https://i.creativecommons.org/l/by-nc-sa/4.0/88x31.png" /></a>104 105## Star History106 107[](https://www.star-history.com/#Junjue-Wang/DisasterM3&Date)108