This public dataset was developed for the PlumeDEBuG project, a data-informed modeling framework for synthetic bubble plume image generation. The dataset contains cropped bubble images collected from real experiments and their corresponding segmentation masks. All bubble samples were manually screened to ensure data quality and to preserve realistic bubble morphologies. By providing experimentally grounded and carefully selected bubble instances, this dataset supports the generation of synthetic bubble images with realistic shapes, enabling the creation of large-scale training data for bubble detection, segmentation, and related image analysis tasks in multiphase flow research.