Research

Text Box: Research

Text Box: 3D reconstruction (height from reference plane) using vertical shadows.
The example above shows a remote reconstruction using a single web cam.

Text Box: A method for synopsis of animation data.
The left example shows a synopsis of motion clip The right example illustrates an application of the method, that is generating thumbnails.

Text Box: Visual appearance representation that captures similarities between different image regions.
It is represented using a generative model that can be learned from a small set of class examples.

Text Box: A frame work for sharing video annotation between to versions of a the same TV program, but of different: format, compression and even insertion and deletion of substantial portions (e.g., a few min. of commercials)

Text Box: Matching 2 video sequences using trajectories of moving objects.
The example above shows recovery of epipolar geometry when 2 cameras are facing each other

Text Box: Feature-Based Sequence-to-Sequence matching

Text Box: Generating high resolution high-frame rate video from multiple unsynchronized low resolution low frame rate video streams.
*Received the Best Paper Prize at ECCV'2002

Text Box: Increasing Space-Time Resolution in Video

Text Box: My first paper on sequence to sequence alignment
See also my Ph.D thesis.

Text Box: Direct Sequence-to-Sequence Alignment

Text Box: Matching non-overlapping sequences using their joint motion.
*Received the Honorable Mention for the 2001 Marr Prize.

Text Box: Alignment of Non-Overlapping Sequences

Text Box: Inducing a segmentation from a single segmented example of image with similar content.
The segmentations on the left of the top left image are induced to the two other test images.

Text Box: A method for generating visual video summarize.
By projecting selected key-poses after rotation, we can visualize self-occluding activities as dynamic still (above) or clip trailer.