Self-Hostable YOLO11 Service for Marine Fish Data Curation

Abstract

Marine fisheries sustainability requires accurate, real-time data on fish populations. This need is currently unmet due to the high costs of manual surveys and the slow, third-party monitoring solutions. This paper introduces a novel self-hostable web service designed for efficient aquatic monitoring and curation of YOLO format datasets. Utilizing FastAPI for asynchronous processing and a YOLO11n architecture, this service enables rapid and high-accuracy fish detection, streamlining the creation of new training data. The self-reliant design eliminates reliance on third-party cloud annotation APIs, addressing privacy concerns and reducing long-term costs associated with data management. The YOLO11n model’s performance was benchmarked against YOLOv10n and YOLOv12n on the DeepFish and OpenImage V7-Fish datasets. It achieved superior detection on DeepFish with an mAP@50 of 97.2% and mAP@50:95 of 61.0%, demonstrating the robust capabilities crucial for effective fisheries management decisions. This work represents an improvement in autonomous marine monitoring technology and promotes data independence. Code is available at: https://github.com/jwuntu/YOLO_MFDDC.