The vision of this project is to make a leap forward in the research on deep learning by making these autonomous deep cognitive models useful for image-based applications anywhere – also for applications on non-standard images with sparsely labelled data. The aim is to arrive at generic models and exploit them to solve key big data computer vision challenges in the marine sector.

 

To fulfil the vision, the project aims to develop cognitive and generic methods that enable autonomous analysis of non-standard, sparsely labelled image data by building on concepts from cross-domain transfer learning, semi-supervised learning and data simulation. More specifically, our main objectives are to:

The project focuses on aiding the automation of three corner stones in computer vision, classification, localization and segmentation. Achieving this will help innovating the next generation big data services for the marine sector.