Computer Science Seminar: Learning Models of Shape and Appearance

Speaker: Dr. Neill Campbell, U. Bath
UCL Contact: lisa.howard@ucl.ac.uk (Visitors from outside UCL please email in advance).
Date/Time: 27 Apr 15, 09:30 - 10:30
Venue: Chadwick B.05

Abstract

As we might imagine, the portrayal of shape and appearance is pervasive throughout the discipline of computer graphics. My current research involves the application of machine learning and computer vision to graphics, with particular interest in the application of probabilistic approaches to modelling 2D and 3D shape and appearance; in this talk I will look at my past, present and future research into this area.

We will start by looking at my early data-driven research into 3D segmentation and reconstruction from many photos. Whilst this work is still state-of-the-art, we observe the need of data-driven approaches to make use of a number of redundant views. When such redundancy breaks down we face a much harder problem; this motivates the need for interactive and learning approaches and I will present an overview of recent work involving user interaction to improve the situation.

My long-term goal, however, is to make use of learning techniques to build strong priors on shape and appearance. In the second part of the talk I will present recent and ongoing research that attempts to learn models and representations of shape automatically for both man-made shapes, such as fonts, and natural shapes, such as animal silhouettes. I will also show how the outputs of such learning can be used to produce intelligent systems that make it easier to process, create and manipulate shapes. Learning approaches can also be extended to appearance and I will illustrate this with demonstrations of appearance transfer and structured inpainting in highly complex, deformable shapes.

Finally, I will provide an overview of some of the future directions for this research, including a near-term project that has just received funding and a longer-term proposal.

Dr. Neill Campbell

Neill is a Lecturer in Computer Science at the University of Bath working in the Graphics and Vision group and also associated with the Centre for Digital Entertainment. Previously, he was a Post Doctoral Researcher at UCL, where he retains a visiting position, working with Jan Kautz and Simon Prince on generating photo-realistic images. He completed his MEng and PhD in the Department of Engineering at the University of Cambridge with funding from a Schiff Scholarship and Toshiba Research. His research interests include: generative models for shape and appearance, multi-view stereo and segmentation, optimisation, active learning, and the application of machine learning techniques to computer graphics and vision problems.