FWISD: The Flood and Waterfront Infrastructure Segmentation Dataset and model performance evaluations

The Flood and Waterfront Infrastructure Segmentation Dataset (FWISD) is constructed for post-disaster assessment, specifically focusing on the impact of Hurricane Francine (September 2024). This dataset utilizes high-resolution UAV imagery to enable precise semantic segmentation of floodwaters, infrastructure damage, and environmental elements. The dataset contains images with a resolution of 1024 x 1024 pixels, sourced from NOAA UAV Imagery, totaling approximately 4.36 GB. The data collection centers on Hurricane Francine during the 2024 Atlantic hurricane season. On September 11, 2024, a Category 2 hurricane struck the southern Louisiana coast, causing widespread flooding and infrastructure threats. This study utilized UAV imagery released by the U.S. National Oceanic and Atmospheric Administration (NOAA) collected between September 16 and 17, 2024, covering multiple severely affected areas in southern Louisiana. To construct a high-quality segmentation dataset, 12 target categories were defined, including natural environmental elements, man-made infrastructure, and disaster-specific elements. The pixel values represent the following classes: 0#Background, 1#Natural Water, 2#Tree, 3#Road-Passable, 4#Road-Flooded, 5#Building-Intact, 6#Building-Damaged, 7#Waterfront Structure-Intact, 8#Waterfront Structure-Damaged, 9#Vehicle-Land, 10#Vehicle-Water, and 11#Floodwater. We designed a standardized pipeline to ensure pixel-level labeling accuracy using LabelMe software. A multi-round iterative quality control mechanism was implemented, including standardization, iterative review, and final inspection to ensure sharp boundaries and accurate class assignments. The dataset follows a standard semantic segmentation directory structure where images and masks are matched by filenames.