Research Projects

 page under construction...

Main topics: Computer Vision and Machine Learning.
Recent projects: Video understanding (text & images), Image segmentation and recognition, approximate inference for Graph Matching and MRFs

Current Research

Video and Text Alignment

Related publications:

Timothee Cour, Chris Jordan, Eleni Miltsakaki, Ben Taskar. Movie/Script: Alignment and Parsing of Video and Text Transcription. Proceedings of the European Conference on Computer Vision (ECCV), 2008.

[videos: retrieval of actions and character naming] 

Approximate inference

Related publications:

Timothee Cour, Jianbo Shi. Solving Markov Random Fields with Spectral Relaxation. Eleventh International Conference on Artificial Intelligence and Statistics (AISTATS), 2007.

Timothee Cour, Praveen Srinivasan, Jianbo Shi. Balanced Graph Matching. Advances in Neural Information Processing Systems (NIPS), 2006

Object specific segmentation and pose estimation

Related publications:

Timothee Cour, Jianbo Shi. Recognizing objects by piecing together the Segmentation Puzzle. IEEE International Conference on Computer Vision and Pattern Recognition (CVPR), 2007.

Normalized Cuts Segmentation Results obtained on the Berkeley Segmentation Dataset
Normalized Cuts segmentation results  obtained on the Berkeley Segmentation Dataset
Multiscale segmentation

We present a multiscale graph-based image segmentation algorithm. In contrast to most multiscale image processing, this algorithm works on multiple scales of the image in parallel, without iteration, to capture both coarse and fine level details. We demonstrate that large image segmentation graphs can be compressed into multiple scales capturing image structure at increasingly large neighborhood. The algorithm has O(N) time complexity, allowing to segment large images with typically N = 1000 x 1000 pixels.

Related publication:

Spectral Segmentation with Multiscale Graph Decomposition. Timothee Cour, Florence Benezit, Jianbo Shi.
IEEE International Conference on Computer Vision and Pattern Recognition (CVPR), 2005.

Learning image segmentation and recognition

We present a general graph learning algorithm for spectral graph partitioning, that allows direct supervised learning of graph structures. Learning is based on gradient descent in the space of graph weights, using derivatives of eigenvectors. This algorithm effectively learns a graph capable of memorizing and retrieving multiple patterns given noisy inputs. We experimented on segmentation and recognition tasks, including bottom-up geometric shape extraction with top-down priors, and hand-written digit recognition.

Related publications:

Learning spectral graph segmentation. Timothee Cour, Nicolas Gogin, Jianbo Shi.
Tenth International Workshop on Artificial Intelligence and Statistics (AISTATS), 2005.

A learnable spectral memory graph for recognition and segmentation. Timothee Cour, Jianbo Shi.
Technical Report MS-CIS-04-12, University of Pennsylvania CIS Technical Reports, Philadelphia, PA, June 2004.

Past projects: Graduate and Undergraduate Research
noisy digit learning
Shape-based object segmentation

internship at University of Pennsylvania, GRASP Laboratory, (2003)
Jianbo Shi

Designed an algorithm to segment objects using shape cues and pixel cues, combined into a consistent grouping of contour and region.



Image Compression and Super-Resolution using Wavelets   [pdf]

Ecole Polytechnique, Department of Applied Mathematics (2002)
Dr. Albert Cohen

Developed an original learning algorithm of super-resolution to optimize the wavelet compression. Increased compression rate over classical linear predictions.


A chess engine and chess-playing autonomous robot   [pdf]

Ecole Polytechnique, Department of Computer Science (2002)
Supervisor: Dr. Eric Goubault

Digital camera, robot and chess engine controlled by a gui. Actual games played by the robot against a human opponent. Robust recognition of chess position under various conditions of light and board position. Used an alpha-beta pruning algorithm with various heuristics

A biologically-inspired neural network for handwritten character recognition   [pdf] [slides]

Ecole Polytechnique, Department of Computer Science (2001)
Pr. Gerard Dreyfus

Back to my home page