Tips for AI-based Image Processing

Based on the experience matured by the iMagine use cases over the first two years, we have a document offering an overview of good practices on AI-based image analysis.

The first part of the document is structured by the main areas of work related to AI-powered image analysis:

  • neural networks for image and video analysis;
  • annotation of images for constructing training datasets;
  • open publication of training datasets; data preprocessing methods prior to AI model training; evaluation metrics and experiment tracking tools;
  • FAIR-ness aspects; recognising and preventing data bias; sharing trained AI models for interference.

The second part of the document provides insight into the eight iMagine use cases and shows how they applied the aforementioned techniques to real scientific scenarios. 

While the document was written from the perspective of aquatic sciences, it is a valuable reading and handbook for any scientific discipline facing challenges in AI-based image analysis.