MSR Vision Faculty Summit - Machine Learning for Visual Recognition

Опубликовано 12 августа 2016, 0:01

This talk begins with a quick overview of machine learning techniques we study for visual recognition. Challenges occur due to high-dimensional space and significant intraclass data variations, which demand good generalisation to unseen data. Among state-of-the-art techniques, we emphasise Randomised Decision Forests and tree-structured methods. Following concepts and principles, their applications are demonstrated for challenging novel problems: real-time action recognition, object phenotype recognition using 3D shape priors, and video-based object recognition, that we recently tackled at Imperial College jointly with Univ. of Cambridge.