News

  • (2016/09) Source code for "deep relative attributes" paper is open-sourced (github).
  • (2016/08) Our paper on deep relative attributes has been accepted to ACCV 2016.
  • (2015/12) A paper on relative attributes submitted to a top tier computer vision conference (available on arXiv).
  • (2015/10) A paper on Persian handwritten digit recognition accepted as an oral to MVIP 2015.
  • (2015/05) A paper on a fast method for localizing bird parts accepted to CVPR 2015 Workshop FGVC3.

Profile

I have recently finished my Master of Science degree at department of computer engineering, Sharif University of Technology where I was a member of IPL under supervision of Prof. Shohreh Kasaei. I also had the privilege to work closely with Dr. Erik Rodner for my thesis. I finished my Bachelors of Science at school of computer engineering, Iran University of Science and Technology where I worked with Dr. Soryani and Dr. Rahmani for my undergrad thesis.

My research so far has been focused on computer vision and machine learning, specifically visual recognition and deep learning.

CV available upon request.

Publications

(Google Scholar)

Deep Relative Attributes

Yaser Souri, Erfan Noury, Ehsan Adeli

Poster, ACCV 2016

arXiv | code

Persian Handwritten Digit Recognition by Random Forest and Convolutional Neural Networks

Yasin Zamani, Yaser Souri, Hossein Rashidi, Shohreh Kasaei

Oral, Iranian Conference on Machine Vision and Image Processing (MVIP) 2015

pdf | ieee link

Fast Bird Part Localization for Fine-Grained Categorization

Yaser Souri, Shohreh Kasaei

The Third Workshop on Fine-Grained Visual Categorization (FGVC3) in conjunction with CVPR 2015

pdf | poster | code

Education

Sep 2013 - Mar 2015

M.Sc. Computer Engineering, Artificial Intelligence

Sharif University of Technology

Under supervision of Prof. Shohreh Kasaei and Dr. Erik Rodner

Thesis: Fine-grained image classification

Sep 2007 - Sep 2011

B.Sc. Computer Engineering, Software Engineering

Iran University of Science and Technology

Under supervision of Dr. Soryani and Dr. Rahmani

Thesis: Facial expression recognition

Projects

Deep Relative Attributes

Deep Relative Attributes

Relative Attributes are a very natural way of thinking in terms of attributes and communicating with machines. The idea was introduced in the award winning ICCV 2011 paper by D. Parikh and K. Grauman. In this project we want to improve their system by using a Deep Neural Network instead of a RankSVM to do the ranking. This way we can also use Convolutional Layers to learn the features end-to-end.

Code: Lasagne, Theano, Python

Collaborators: Erfan Noury, Ehsan Adeli

Bird Part

Fast Bird Part Localization (Fine-grained recognition)

A fast way to detect bird parts (e.g. head) using pre-pixel deep features and random forests is introduced. Then using the detected part locations we can do fine-grained recognition on CUB-200-2011. Our mean accuracy is 72.02% which is comparable with state-of-the-art (73.89%) while being at least 2 orders of magnitude faster when detecting parts. Our method only uses a single forward pass of the network to detect parts.

Code: Python, Caffe, Scikit-learn

Collaborators: Shohreh Kasaei, Erik Rodner

Field Lines

Supervised Detection of Field Lines

An annotator in flask and javascript. Plus the edge detection code of Piotr Dollar.

Code: Matlab, Python, Javascript

Collaborator: Erfan Noury (He made a much more sophisticated version of the annotator, which is available here)

Mahv

Mahv

An original implementation of Criminisi's image impainting algorithm for a course project.

Code: OpenCV and C++

Misc

  • Academic countdown
  • Abstracts: A simple interface to quickly browse papers accepted into computer vision conferences. (Highly Opinionated)