PhD Opportunities

  • The group considers open applications for 3.5 year PhD positions from high quality candidates who satisfy RCUK residency requirements (the departmental details are available here). For potential project ideas, please read on.
  • Unfortunately, I'm afraid I do not have funding or space to host interns so please do not ask.

Potential PhD Projects

Please see the bottom of the page for further details on requirements and applying.

Learning Shape and Appearance Models of Deformable Objects

The core themes of our group's research are applying Machine Learning techniques to solve problems in Computer Vision and Computer Graphics. This new project looks to extend our recent work in learning manifold models of shape into learning generative models of shape and appearance for a variety of tasks.

For reference please see our latest SIGGRAPH paper that applies these techniques to fonts (as a class of shape); there is an interactive demo at the following link: http://bit.ly/FontManifold (please be patient, the site takes a while to load).

We have begun to expand these approaches to other categories of shape; e.g. the following is an example shape manifold for elephants used in our paper on interactive sketch-based image synthesis:

We plan to advance these approaches and build new models to learn automatically the shape and appearance of a range of objects, mostly involving complex deformable models.

A few of the many applications include:

  • Visual effects industry: rotoscoping (segmentation) and motion capture
  • Vision: tracking, recognition and reconstruction of deformable objects
  • Graphics: artist assistance; editing and generation of photo realistic objects
  • Sports and Bio-sciences: detailed tracking and capture of human and animal movement
  • Visual Assistance: navigation and scene interpretation for those with visual disabilities

For these examples we either have interested industrial contacts or contacts in other departments at the University of Bath so we can make sure to get the results out into the real world as quickly as possible.

Multi-View Stereo

Multi-View Stereo (MVS) is concerned with recovering 3D models of objects and surfaces from a set of images (or video) taken from different viewpoints. An example result using many images of a vase in a museum is:

Whilst this area has received a lot of attention and is becoming a mature technology (for example Google/Apple/Bing Maps), there are still many areas with no solutions or poor quality. I am interested in extending and developing new MVS techniques to deal with the following challenges:

  • Deformable surfaces

    where the usual assumption of a static and rigid object or surface is violated (e.g. a moving face or a walking animal)

  • Semantic reconstruction

    where the category and shape of an object/surface are estimated simultaneously and used to improve the quality of both

  • Learning approaches

    the majority of MVS methods make use of heuristically designed energy functions and regularisation; instead we would prefer to learn such models from training data and condition them on observations at test time

Industrial Contacts

We have access to a range of industrial contacts that will be available to interested students. These include: The Foundry, The Imaginarium, Double Negative, Adobe, nVidia and Google. There is also the chance to go on a paid internship to work in a research department in the same field for students who demonstrate sufficient ability and motivation; these are intended to lead to further collaboration and shared authorship of papers.

Requirements

The University requires students to have a first or upper second class honours degree with some further requirements for non-native English speakers (details available here).

Candidate without a Masters level course in computer vision, computer graphics, machine learning, applied mathematics, or a strongly correlated field would have to provide strong justification (and evidence) that they would be able to handle the maths and programming necessary to complete a PhD in this field.

Programming experience is a particular advantage, specifically proficiency in on or more of C++ / Python / Matlab. All of the techniques we use build on Linear Algebra and it would be desirable for the candidate to have some experience in applied mathematics / numerical methods.

Applying!

If you meet the requirements above and are interested in vision / graphics / machine learning then please get touch by contacting me and putting "PHD POSITION" in capitals at the start of the subject line of the email; please include a copy of your CV and feel free to include your motivations for doing a PhD and any questions you would like to ask me. I very much look forward to hearing from you.