Researchers have released a new benchmark dataset designed for active-fire segmentation using Sentinel-2 satellite imagery. This dataset comprises 2,148 image-mask pairs derived from 25 California wildfires, covering periods from July 2020 to August 2026. The data includes detailed masks distinguishing background, active fire, and invalid observations, along with associated metadata and code for training and evaluation. The dataset aims to support research in rare-class segmentation and learning from algorithmic labels. AI
IMPACT This dataset could advance research in rare-class segmentation and improve AI models for wildfire detection and monitoring.
RANK_REASON The cluster describes the release of a new benchmark dataset for a specific research task (active-fire segmentation) based on satellite imagery, accompanied by code and metadata. [lever_c_demoted from research: ic=1 ai=1.0]
- August 2026
- California
- GitHub
- July 2020
- Level-2A
- Mohammadreza Narimani
- ResNet-34 U-Net
- Sentinel-2
- Zenodo
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →