Omid Aghazadeh, Ph.D.

Education:

• Ph.D in Computer Vision
  April 2011 - June 2014/ Computer Vision and Active Perception Laboratory, Royal Institute of Technology (KTH), Stockholm, Sweden
• M.Sc in Systems, Control and Robotics
  Sept 2008 - June 2010/ Royal Institute of Technology (KTH), Stockholm, Sweden
• B.Sc in Computer Engineering(Software)
  2001 - 2006 / Azad University of Qazvin

Publications:

Conference Papers:

• Estimating Attention in Exhibitions Using Wearable Cameras/ A. S. Razavian, O. Aghazadeh, J. Sullivan, S. Carlsson, ICPR, 2014.
• Properties of Datasets Predict the Performance of Classifiers / O. Aghazadeh, S. Carlsson, BMVC, 2013. [PDF] , [Poster] , [Extended Abstract] , [Supp Mat]
• Mixture Component Identification and Learning for Visual Recognition/ O. Aghazadeh, H. Azizpour, J. Sullivan and S. Carlsson, ECCV, 2012. [PDF] , [Poster] , [Supp Mat] , [ECCV Demo] , [Clustering Results] , [Trained Models for Pascal VOC 2007]
• Multi View Registration for Novelty/Background Separation/ O. Aghazadeh, J. Sullivan and S. Carlsson, CVPR, 2012. [PDF] , [Poster] , [Supp Mat]
• Novelty Detection from an Ego-Centric Perspective/ O. Aghazadeh, J. Sullivan and S. Carlsson, CVPR, 2011. [PDF] , [Poster] , [CEBIT Demo]
• Reading Street Signs Using a Generic Structured Object and Signature Recognition/S. N. Parizi, A. T. Targhi, O. Aghazadeh and J. O. Eklundh, VISAPP, 2009. [PDF]

Workshops / Presentations :

• Datasets, Sample Selection, and Test Performance/ O. Aghazadeh and S. Carlsson, The 1st International Workshop on Visual Domain Adaptation and Dataset Bias (in conjunction with ICCV 2013), 2013. [Extended Abstract] , [Poster]
• Automatic Memory Selection via Novelty Detection/ O. Aghazadeh, J. Sullivan and S. Carlsson, Oxford Sense Cam 2012, 2012. [Abstract] , [Poster]
• Novelty Detection in Wearable Vision Systems/ O. Aghazadeh, J. Sullivan and S. Carlsson, Second IEEE Workshop on Egocentric Vision (in conjunction with CVPR 2012), 2012. [Extended Abstract] , [Presentation]

Non Peer Reviewed Publications / Manuscripts:

• Large Scale, Large Margin Classification using Indefinite Similarity Measures/ O. Aghazadeh and S. Carlsson, 2014. [arXiv]
• Human Pose Estimation from RGB Input Using Synthetic Training Data/ O. Danielsson and O. Aghazadeh, 2014. [arXiv]
• Object Segmentation using Spatial and Spatio-Temporal Cues/ O. Aghazadeh, J. O. Eklundh and J. Sullivan, 2011. [PDF] , [Supp Mat]

Thesis:

• Doctoral thesis: Data Driven Visual Recognition, 2014
• MSc thesis: Object Segmentation Using Spatial and Spatio-Temporal Features, 2010

Selected Projects:

• Visual Diaries:



Project Page

• Foreground/Background Object Segmentation:


• Object Boundary Detection using Spatio-Temporal(Optical Flow) and Visual(Texture and Color) Cues:


• Non-Linear and Anisotropic Image Denoising:



Original Image	Gaussian Homogeneous Smoothing	Gaussian Anisotropic Smoothing
Original Image	Gaussian Smoothing	Edge Enhancing Smoothing (TV Like De-noising)

• Scale-Space Methods:



Gaussian Scale Space	Perona-Malik Scale Space	Total-Variation Scale Space

Teaching Experiences:

• 2009, Teaching Assistant in "Applied Estimation", KTH.
• 2010, Teaching Assistant(one session) in "Applied Estimation", KTH. Introduction to Vectorized Processing in MATLAB , Sample Codes for Vectorized processing in MATALB

Code:

• Iterative methods for computation of Fundamental matrix: The package contains MATLAB implementation of the iterative algorithms suggested in Hartley and Zisserman's Multi View Geometry.
• Nonlinear Coupled Diffusion: The package contains MATLAB implementation of Weickert's Nonlinear Coupled Diffusion.
• A Sparse Set of Features for Texture Discrimination: The package contains MATLAB implementation of Brox's feature set.
• Gaussian Mixture Models on Cuda Mex(Binary) [Requires Fermi Cards]: The package contains linux 64 bit binaries for learning(Expectation Maximization) and evaluating multi dimensional gaussian mixture models using GPU acceleration.
• Isotropic Multi Dimensional Kernel Density Estimation on Cuda Mex(Binary) [Requires Fermi Cards]: The Linux 64 bit binary for MDKDE(Isotropic) using GPU acceleration.
• Dense distance computations between two sets of points(Binary) [Requires Fermi Cards]: The Linux 64 bit binary for dense distance computations including P norms(L_1, L_2 and L_inf), Chi squared distance and Histogram Intersection between two sets of points(or histograms) using GPU acceleration. The code is tuned to handle large dimensions and large sets (when the distance matrix does not fit on device memory).