Thursday 10 June 10:30 Teknikringen 14, Room 304
Recovering body configurations of people big and small
Alexei A. Efros
Robotics Research Group
University of Oxford
In this talk, I will present and contrast two approaches for recovering
human body configurations in two very different recognition regimes
-- medium field (the 30-pixel man) and near field (the 300-pixel
man). In the medium field, where information at each frame is very
poor, we use motion to recognize a set of human actions in a pure
exemplar-based approach. Once the actions have been matched, body
configurations can be "transfered" from the training data
onto the novel figure. This simple method is also applied to motion
retargetting: synthesizing a novel person imitating the actions
of another person ("Do as I Do" synthesis) or performing
actions according to the specified action labels ("Do as I
Say" synthesis). In the near field, where much more information
is available in a still image, we attempt to tackle the body recovery
problem in a general setting. The dataset we use is a collection
of sports news photographs of baseball players, varying dramatically
in pose and clothing. The approach that we take is to use segmentation
to guide our recognition algorithm to salient bits of the image.
We use this segmentation approach to build limb and torso detectors,
the outputs of which are assembled into human figures.
This is joint work with Greg Mori, Xiaofeng Ren, Alex Berg, and
Jitendra Malik at UC Berkeley.
Friday 11 June, 10:30, Teknikringen 14, Room 304
Visual Recognition from Invariant Local Features
Computer Science Department
University of British Columbia
Within the past few years, invariant local features have been successfully
applied to a wide range of recognition and image matching problems.
For recognition applications, it has proved particularly important
to develop features that are distinctive as well as invariant, so
that a single feature can be used to index into a large database
of features from previous images. Robust recognition can then be
achieved by identifying clusters of features using a Hough transform
and detailed model fitting. Recent work will be presented on applications
of this approach, including location recognition and the detection
of image panoramas from unordered sets of images.