Earthquake Infrastructure Damage Semantic Segmentation(EIDSeg) Dataset for Social Media Images, in EIDSeg: A Pixel-Level Semantic Segmentation Dataset for Post-Earthquake Damage Assessment from Social Media Images

The EIDSeg Dataset (Earthquake Infrastructure Damage Segmentation Dataset) is a large-scale, pixel-level annotated dataset designed for semantic segmentation of post-earthquake infrastructure damage using ground-level social media imagery. The dataset provides a unique resource for advancing computer vision and machine learning applications in disaster response and damage assessment. EIDSeg contains 3,266 ground-level images collected from nine major earthquake events between 2008 and 2023, covering a wide range of geographic regions, building types, and damage severities. Each image is manually annotated at the pixel level into five semantic classes that represent key types of infrastructure and damage states: Undamaged Building, Damaged Building, Destroyed Building, Undamaged Road, and Damaged Road. The dataset is organized into train, validation, and test subsets, each containing images and corresponding annotation files in CVAT XML format. EIDSeg provides a standardized resource to support research on fine-grained damage assessment, computer vision for disaster response, and domain-adaptive segmentation models. More details of the paper and code is updated in https://github.com/HUILIHUANG413/EIDSeg.