## Highlights
DisasterM3 includes 26,988 bi-temporal satellite images and 123k instruction pairs across 5 continents, with three characteristics:
1. Multi-hazard: 36 historical disaster events with significant impacts, which are categorized into 10 common natural and man-made disasters
2. Multi-sensor: Extreme weather during disasters often hinders optical sensor imaging, making it necessary to combine Synthetic Aperture Radar (SAR) imagery for post-disaster scenes
3. Multi-task: 9 disaster-related visual perception and reasoning tasks, harnessing the full potential of VLM's reasoning ability
## News
- 2026/11/05, We released the official training code.
- 2026/09/30, 🔥🔥The second version of our disaster benchmark ([DORA](https://github.com/Junjue-Wang/DORA)) got accepted by NeurIPS 2025 as an Oral paper!🔥🔥
- 2025/10/23, We released the DisasterM3 [instruct set](https://forms.gle/APQpmyuThh28HsJdA).
- 2025/10/17, We released the DisasterM3 [benchmark set](https://forms.gle/APQpmyuThh28HsJdA).
- 2025/09/22, We are preparing the dataset and code.
- 2025/09/22, Our paper got accepted by NeurIPS 2025.
## Benchmark
Please run this code for benchmarking the DisasterM3 dataset.
First prepare the data. Download the benchmark set from [Hugging Face](https://huggingface.co/datasets/Kingdrone-Junjue/DisasterM3) or the [form](https://forms.gle/APQpmyuThh28HsJdA), then write one file per subset to `data/` next to the repository folder:
```
git clone https://github.com/Junjue-Wang/DisasterM3.git disaster_m3
python disaster_m3/pyscripts/split_benchmark.py --bench_root /path/to/DisasterM3_Bench
```
`--bench_root` is the folder that holds `benchmark_release.json` and `test_images/`.
`PYTHONPATH` must point to the repository folder:
```
export PYTHONPATH=$PWD/disaster_m3
```
Two examples:
Qwen2.5 VL:
```
python disaster_m3/pyscripts/run_vllm.py --model_id Qwen/Qwen2.5-VL-7B-Instruct --subset bearing_body
```
InternVL3:
```
python disaster_m3/pyscripts/run_vllm.py --model_id OpenGVLab/InternVL3-78B --subset recovery
```
## Training
The official training code is in [training/](training/README.md). It covers LoRA fine-tuning of Qwen2.5-VL-7B and InternVL3-8B with LLaMA-Factory, merging the weights, and running the merged models on the benchmark.
## Citation
If you use DisasterM3 in your research, please cite our following papers.
```text
@inproceedings{wang2026dora,
title={Can LLM Agents Respond to Disasters? Benchmarking Heterogeneous Geospatial Reasoning in Emergency Operations},
author={Wang, Junjue and Xuan, Weihao and Qi, Heli and Dai, Pengyu and Liu, Kunyi and Chen, Hongruixuan and Zheng, Zhuo and Xia, Junshi and Ermon, Stefano and Yokoya, Naoto},
booktitle={Proceedings of the Neural Information Processing Systems},
year={2026}
}
@inproceedings{wang2025disasterm3,
title={DisasterM3: A Remote Sensing Vision-Language Dataset for Disaster Damage Assessment and Response},
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},
booktitle={Proceedings of the Neural Information Processing Systems},
year={2025}
}
```
## Acknowledgments
This dataset builds upon the following excellent open datasets:
- **xBD dataset** by Ritwik Gupta
- [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)
- [Dataset](https://xview2.org/dataset)
- License: [CC BY-NC-SA 4.0]
- **BRIGHT dataset** by Hongruixuan Chen
- [Repository](https://github.com/ChenHongruixuan/BRIGHT)
- License: [CC BY-NC 4.0]
## License
All images and their associated annotations in DisasterM3 can be used for academic purposes only,
but any commercial use is prohibited.
## Star History