Deep Visual Learning

Vision

The Deep Visual Learning (DVL) Lab conducts research in the areas of deep learning and computer vision. Our work focuses on developing novel methodologies for learning models to automatically analyze images and videos. We are especially interested in developing deep models which successfully adapt and continually learn from real-world visual data. Recent projects include object recognition and detection, pixel-level prediction tasks (e.g. depth estimation from monocular input, semantic segmentation) and human behaviour analysis and synthesis.

Contributions to Smart Cities and Communities

The unit contributes to the activities of Smart Cities and Communities through research and innovation in the area of automatic video analysis for smart city sensing. DVL develops algorithms for people detection, tracking and human behaviour understanding from visual streams which are key components of video surveillance systems. Furthermore, information gathered from visual input requires to be integrated with other data sources (e.g. audio, mobility data) and DVL researchers have a long standing experience in working with multi-modal data.

Contributions to Artificial Intelligence

DVL works at the forefront of computer vision and artificial intelligence by developing novel deep learning models for addressing several challenging tasks involving the automatic processing of visual data. Our recent research efforts are devoted to design and validate novel deep architectures for incremental and continual learning and for robust adaptation in presence of visual appearance changes. Besides methodological contributions, DVL researchers also focus on applying the learned models in real-world applications (e.g. video surveillance, human robot interaction). DVL is also involved in the new Master Degree in Artificial Intelligence Systems at University of Trento.