In 2018 I worked as a Computer Vision Researcher under Professor Kaleem Siddiqi and Morteza Rezanejad in the Shape Analysis Group at the McGill Centre For Intelligent Machines. Our project combined research in human and computer vision to study the role of contour geometry in scene categorization. By leveraging ideas from human psychology and perception such as symmetry and gestalt grouping we were able to enhance performance of state-of-the-art computer vision models for classifying line drawings of scenes.

The paper I collaborated on for the project has now been accepted to CVPR 2019! You can find our paper on arxiv or on the official CVPR site. This was, as far as we know, the first study of classification of line drawings of scenes by Convolutional Neural Networks (CNNs)

The code we wrote for generating line drawings of scenes is available on GitHub. I wrote the command line interface and documentation for the repo. The matlab code and algorithms were developed by Morteza Rezanejad and Professor Kaleem Siddiqi.

You can also find all of the code for the machine learning experiments used in the paper on my GitHub. I wrote all of the experiments using Keras and TensorFlow.