A high-resolution beach imagery dataset with COCO annotations for deep learning-based Sargassum monitoring in coastal environments.

Accurate monitoring of Sargassum blooms is essential for mitigating their severe socioeconomic and ecological impacts on coastal areas. To support the development of computer vision tools for coastal management, this study presents a dataset of high-resolution images along the coast of Quintana Roo, Mexico. The dataset comprises RGB images, accompanied by pixel-level multi-class segmentation masks. These masks classify the coastal environment into three distinct classes: Sargassum, sand, and other environmental elements and coastal objects (such as water, vegetation, sky, clouds, boats, and people, among others). To facilitate immediate integration with modern deep learning architectures, the pixel-level masks have been systematically converted into polygonal annotations formatted as COCO-compatible JSON files. The dataset also includes descriptive statistics detailing class distribution to characterize spatial complexity. This curated, multi-format dataset provides a standardized reference baseline for researchers and developers with which to train, validate, and compare semantic and instance-level segmentation models (e.g., U-Net, Mask R-CNN) aimed at automated quantification of Sargassum biomass and efficient coastal cleanup logistics. This dataset goes beyond the previously published collection, making it the current benchmark of its kind.