Karim Ali Karim Ali

Algocian
Phone: (+1) 514 574 2686
E-mail: karim.ali=[at]=algocian.com

Research

I am currently CEO at Algocian, a Computer Vision startup focusing on Automotive and Security applications.

Prior to that, I was a Postdoctoral Fellow at the University of California, Berkeley and the University of Massachussets, Lowell, as well as a Visiting Scholar at Harvard Univeristy. My interests lie in the application of statistical machine learning to vision problems.

I am committed to developing algorithms whose outputs in real world scenarios are sufficiently reliable to be of practical use.

  • Learning Context Cues for Synapse Segmentation
    Electron Microscopy data poses unique challenges for automatic segmentation algorithms in part because the volumes are heavily cluttered with structures that exhibit very similar textures and are difficult to distinguish solely on the basis of local image statistics. We propose an approach designed to take such contextual cues into account and emulate the human ability to distinguish regions that merely share a similar texture. Our method processes the data directly in 3D and is specifically designed to leverage context cues. We demonstrate our ability to automatically process large EM stacks, reliably collect density, shape and orientation statistics from thousands of synapses.
tools
  • Data-Driven Visual Tracking in Retinal Microsurgery
    In the context of retinal microsurgery, visual tracking of instruments is a key component of robotics assistance. The difficulty of the task and major reason why most existing strategies fail on in-vivo image sequences lies in the fact that complex and severe changes in instrument appearance are challenging to model. We introduce a novel approach, that is both data-driven and complementary to existing tracking techniques. In particular, we show how to learn and integrate an accurate detector with a simple gradient-based tracker within a robust pipeline which runs at framerate.
tools
  • FlowBoost
    It is now possible to build robust object detectors by inputing a sufficient amount of labeled data to highly generic black-box algorithms (Boosting, SVM, Decision Trees, Neural Networks, ect.). A main drawback of these methods is that such labeled data needs to be manually collected. We propose a framework to significantly reduce the amount of labeled data to be collected by exploiting the temporal consistency occurring in a training video. Our method, FlowBoost, allows for a reduction in the number of training frames up to a factor of 60. This comes with virtually no loss in performance and in some instances, gains in performance are observed.
cells
  • A Real-Time Deformable Detector
    One main challenge in object detection is that object appearance can change signifcantly depending on the pose of the object. This problem is typically delt with using a view-based approach whereby a series of detectors, each specialized to a particular viewpoint, are built. We present a framework that enables a single detector, whose constituent features deform, to achieve reliable detection. As an added benefit, our framework forgoes the need to annotate training data for pose and is applicable to a variey of machine learning methods such as Boosting and Support Vector Machines.
Cars

Publications

X. Peng, B. Sun, K. Ali and K. Saenko
IEEE International Conference on Computer Vision, 2015.
X. Peng, B. Sun, K. Ali and K. Saenko
International Conference on Learning Representations (Workshop track), 2015.
K. Ali and K. Saenko
IEEE Conference on Computer Vision and Pattern Recognition, 2014.
C. Becker, K. Ali, G. Knott and P. Fua
PREPRINT - IEEE Transactions on Medical Image Analysis, 2013
K. Ali, F. Fleuret, David Hasler and P.Fua
IEEE Transactions on Pattern Analysis and Machine Inteligence, 2012
C. Becker, K. Ali, G. Knott and P. Fua
In Medical Image Computing and Computer Assisted Intervention Conference, 2012.
R. Sznitman, K. Ali, R. Richa, R. Taylor, G. Hager and P. Fua
In Medical Image Computing and Computer Assisted Intervention Conference, 2012.
K. Ali, D. Hasler and F. Fleuret
IEEE Conference on Computer Vision and Pattern Recognition, 2011.
C. Stretcha, A. Lindner, K. Ali and P. Fua
In 31st DAGM Symposium on Pattern Recognition, 2010
K. Ali, F. Fleuret, David Hasler and P.Fua
IEEE International Conference on Computer Vision, 2009
K. Ali and F. Labeau
IEEE Vehicular Technology Conference, 2005
K. Ali and F. Labeau
IEEE Canadian Workshop on Information Theory, 2005

Bio

Ph.D., Computer and Communication Sciences, EPFL. 2007-2012
Analyst, Business and System Integration Consulting, Accenture Ltd. 2005-2007
M. Eng, Telecommunications and Signal Processing, McGill University. 2002-2005
B. Eng, Honors Electrical Engineering, McGill University. 1998-2002

Dissertations

Ph.D. Thesis, 2012.
Master's Thesis, 2005
Honors Thesis, 2002

Misc

  • Curriculum Vitae
  • My Ph.D advisors: Dr. Fran├žois Fleuret and Prof. Pascal Fua
  • I worked at the CSEM on training of embedded visual systems with Dr. David Hasler
  • LinkedIn