Exploring Bubble Flow with Deep Learning: A Preliminary Study

In a bubble flow study, the bubble size measurement requires identifying every instance in each im-age frame in the first place. However, their tininess and highly overlapped instances hindered notonly human-level manipulation but also any deep learning-based detectors. In addition, their densepresentation in the observation zone made manually measuring phase time-consuming and labo-rious. Annotation pipelines, adopted foundation models, and human intervention were proposed;those accelerate this process and help to build a real-world bubble detection dataset ready to fine-tune a deep learning model faster. Experimental results yield YOLO as the best balancing detectorwhen coupled with upscaling during training for tiny bubble detection and its application in bubblesize distribution