iMagine: Bridging AI and image processing in aquatics sciences

iMagine [1] offers cutting-edge machine learning solutions for aquatic sciences, providing AI-driven tools and methodologies to enhance the processing and analysis of images and videos in marine and freshwater research. By supporting the study of oceans, seas, coasts, and inland waters, it contributes to a deeper understanding of aquatic health. As part of the European Open Science Cloud (EOSC), iMagine facilitates the development, training, and deployment of AI models.

Through lessons learned from the collaboration between iMagine and eight initial use cases, along with six additional use cases onboarded via open calls, all spanning various aquatic science fields, a comprehensive collection of best practices has been compiled and published. These best practices focus on analyzing aquatic science imagery using AI, covering the entire pipeline from data preparation to model deployment as a service and monitoring.

The platform ensures labeling consistency and accuracy by providing well-defined annotation guidelines and verified tools, like CVAT. Users can access federated computing resources via a web browser for developing and training AI models. ... mehriMagine offers a marketplace [2] for sharing models, complemented by rich metadata and an option to try each model. The AI models are encapsulated as Docker images and integrated with a REST API (DEEPaaS) to ensure reproducibility, and it simplifies a service provision. The platform also facilitates the deployment of inference services using an open-source serverless computing framework (OSCAR). Researchers benefit from the platform’s flexibility, allowing seamless execution of AI models on federated clouds. The platform utilizes MLflow for experiment tracking, enabling transparent and reproducible AI model evaluations.

Additionally, iMagine addresses challenges such as drift detection and data FAIRness, providing case studies on AI’s limitations and advancements in aquatic sciences. By integrating the FAIR EVA tool, iMagine improves dataset reliability and facilitates FAIR data publishing in repositories like Zenodo.

In this work, we present the iMagine platform, highlighting AI-driven tools and best practices for the complete aquatic imagery analysis pipeline.

References:
[1] https://imagine-ai.eu
[2] https://dashboard.cloud.imagine-ai.eu/marketplace