_____________________________________________________________________
SUMMARY
The project aims at exploiting large collections of unlabeled multi-modal data, mainly video footage, to further the development of state-of-the-art technology in video, audio and natural language understanding, interpretation, annotation and retrieval by combining unsupervised and semi-supervised learning. It will address problems that are very difficult (some probably impossible) to solve in a single modality by adopting an interdisciplinary approach. Progress in individual research areas – vision, natural language processing and speech will be achieved by co-training and by exploiting the results of other modalities as cross-training data. For the efficient processing of large data collections appearing in all modalities, the project will also concern generic problems of organisation, indexing, and searching based on a similarity that is necessary for building real-life applications.
_____________________________________________________________________
SOLUTIONS
- Designing the technology necessary for semantically consistent human motion segmentation
- Developing a key-pose similarity algorithm for motion data retrieval
_____________________________________________________________________
LINK: http://www.dri.ie/projects
_____________________________________________________________________
