MMFire

MMFire is a wildfire spread dataset, with inputs provided by the LANDFIRE program [1,2], and outcomes simulated using the SimFire [3] software, which internally uses the Rothermel equations [4]. The dataset serves as a benchmark for ambiguous segmentation tasks. For each input condition describing a wildfire and environmental conditions, it contains eight different possible outcomes. For the simulator to produce different outcomes, we vary the wind direction between eight options (North, North-East, East, ... ), set as constant over the whole 64x64 pixel area. Please note that this dataset does not represent realistic wildfire spread. It is only meant as a tool to study ambiguous segmentation. The eight different outcomes per input are not a realistic approximation of future outcomes. However, since reality only provides us with a single outcome, we resort to simulations to be able to validate our methods. Since the relationship between input and outputs is a relatively simple computer program, be aware that models with enough capacity should be able to approximate this relationship very well. If you use this dataset, please cite the corresponding paper: @inproceedings{ gerard2025wildfire, title={Wildfire Spread Scenarios: Increasing Sample Diversity of Segmentation Diffusion Models with Training-Free Methods}, author={Sebastian Gerard and Josephine Sullivan}, booktitle={Northern Lights Deep Learning Conference 2026}, year={2025}, url={https://openreview.net/forum?id=E44d5hzEV0} } [1] LANDFIRE. Anderson Fire Behavior Fuel Model (FBFM13) Layer. LANDFIRE 2.0.0. U.S. Department of the Interior, GeologicalSurvey, and U.S. Department of Agriculture, 2016. url: https://landfire.gov/[2] LANDFIRE. Elevation Layer. LANDFIRE 2.2.0. U.S. Department of the Interior, Geological Survey, and U.S. Department of Agriculture, 2020. url: https://landfire.gov/[3] Doyle, M., Threet, M., Dotter, M., Kempis, C., Tapley, A., & Welsh, T. (2024). SimFire (Version 2.0.1) [Computer software]. https://github.com/mitrefireline/simfire[4] R. C. Rothermel. “A mathematical model for predicting fire spread in wildland fuels”. In: Res. Pap. INT-115. Ogden, UT: U.S. Department of Agriculture, Intermountain Forest and Range Experiment Station. 40 p. 115 (1972). url: https://www.fs.usda.gov/research/treesearch/32533MMFire是一款野火蔓延数据集，其输入数据由LANDFIRE项目[1,2]提供，模拟结果通过SimFire[3]软件生成，该软件内部采用了Rothermel方程[4]。本数据集作为模糊分割任务的基准数据集。针对每一组描述野火及环境条件的输入条件，数据集包含八种不同的可能蔓延结果。为使模拟器生成差异化结果，我们将风向设置为八种可选方向（北、东北、东……），并令其在整个64×64像素的区域内保持恒定。 请注意，本数据集并未还原真实的野火蔓延过程，仅作为研究模糊分割任务的工具。每组输入对应的八种结果并非对真实未来蔓延场景的合理近似。但由于现实中仅能获取单一的实际蔓延结果，我们通过模拟来实现对模型方法的验证。 由于输入与输出之间的映射关系为相对简单的计算机程序，需注意：具备足够容量的模型可以很好地拟合该映射关系。 若使用本数据集，请引用对应论文： @inproceedings{ gerard2025wildfire, title={Wildfire Spread Scenarios: Increasing Sample Diversity of Segmentation Diffusion Models with Training-Free Methods}, author={Sebastian Gerard and Josephine Sullivan}, booktitle={Northern Lights Deep Learning Conference 2026}, year={2025}, url={https://openreview.net/forum?id=E44d5hzEV0} } [1] LANDFIRE. 安德森火灾行为燃料模型（FBFM13）图层. LANDFIRE 2.0.0. 美国内政部、地质调查局及美国农业部, 2016. 网址：https://landfire.gov/ [2] LANDFIRE. 高程图层. LANDFIRE 2.2.0. 美国内政部、地质调查局及美国农业部, 2020. 网址：https://landfire.gov/ [3] Doyle, M., Threet, M., Dotter, M., Kempis, C., Tapley, A., & Welsh, T. (2024). SimFire（版本2.0.1）[计算机软件]. https://github.com/mitrefireline/simfire [4] R. C. Rothermel. “野火燃料中火灾蔓延预测的数学模型”. 载于：研究报告INT-115. 奥格登，犹他州：美国农业部山间森林与牧场试验站. 40页. 115 (1972). 网址：https://www.fs.usda.gov/research/treesearch/32533