ABOUT ME
Hello!
I am a researcher focusing on vision and perception for virtual humans at the Graphics and Media at York (GaMaY) Lab at York University in Toronto, Ontario, Canada, under the supervision of Dr. Petros Faloutsos. I am also an external collaborator with the Graphics, Artificial Intelligence, Design, and Games(GAIDG) Lab at the University of Victoria in Victoria, British Columbia, Canada, as well at the Intelligent Visual Interfaces Lab at Rutgers University in New Jersey, USA.
I am personally interested in integrating visual attention models and techniques into virtual human simulation.
I am also a 3D artist, check out my works!
RECENT WORK
Parametric Saliency Maps for Autonomous Pedestrians
The focus of my masters thesis was a framework for generating "psuedo" saliency maps in Unity (C#). The parametric model is implemented in a shader (HLSL), which incorporates 7 attention factors in order to generate 2D saliency maps in runtime from the perspective of a virtual agent.
Methodologies used: Unity (C#), software prototyping, parameter optimization, usability testing, experiment piloting.
If your browser does not support embedded videos, click here to view on YouTube.
Below are results from optimizing the parametric saliency maps model to more closely resemble SALICON, a state of the art saliency model. Optimization was done using the CMA-ES algorithm. Three loss functions were experimented with across three objective functions (9 objective/loss pairs), based on saliency metrics of SIM score and KL-Divergence which measure the similarity of distributions and are a good co-pair of metrics due to complementary features. Points represent comparisons between pairs of generated parametric saliency maps and generated SALICON saliency maps. Orange points are like-pairs, corresponding to the same visual input. Blue points are unlike-pairs. Optimization resulted in clear strong separability of like-pairs from unlike-pairs, despite the parametric model having only 7 degrees of freedom. Interestingly, even the unoptimized baseline shows some inherent structural differences of like-pairs from unlike-pairs, which suggests the choice of attention parameters themselves encode a good deal of what machine learning trained models are encoding. Further investigation would be needed to confirm this.
.png)
.png)
Tank following Hermite Spline, Rocket Tracking (OpenGL)
Implementation of a hermite curve-driven vehicle in C++ and OpenGL.
If your browser does not support embedded videos, click here to view on YouTube.
Spring Mass Particle System (OpenGL)
Spring mass particle system with ground repulsion forces in C++ and OpenGL. Supports Euler, Symplectic, and Verlet integration.
If your browser does not support embedded videos, click here to view on YouTube.
Inverse Kinematics System (OpenGL)
Implementation of an inverse kinematic system in C++ and OpenGL.
If your browser does not support embedded videos, click here to view on YouTube.
Papers and Reports
Publications
Automatic estimation of parametric saliency maps (PSMs) for autonomous pedestrians
May 2022
Computers & Graphics 104
PSM: Parametric Saliency Maps for Autonomous Pedestrians
November 2021
Motion, Interaction and Games Conference
Graduate Coursework
Evaluating Search and Selection Times for Linear, Grid, and Radial GUI Menus
December 2020
Research Methods for Human Computer Interaction
York University, Toronto, Ontario, Canada
Human Motion State Classification from Motion Capture Data
April 2021
Data Analytics and Visualization
York University, Toronto, Ontario, Canada
Object Feature Mapping and View Prediction from Same-Different Determiniation of 3D Polyhedrons
April 2021
Advanced Topics in Computer Vision
York University, Toronto, Ontario, Canada
Reproducibility Study: Neural Pose Transfer by Spatially Adaptive Instance Normalization
April 2021
Neural Networks and Deep Learning
York University, Toronto, Ontario, Canada
What Do Neural Networks Learn When Trained With Random Labels? (Paper Report)
December 2020
Machine Learning Theory
York University, Toronto, Ontario, Canada
ACADEMICS
Education
MSc - Computer Science - Computer Graphics
September 2020 - September 2022
York University, Toronto, Ontario, Canada
BSc with Honours, Double Major - Computer Science & Physics
September 2015 - April 2020
York University, Toronto, Ontario, Canada
Selected Courses
| Neural Networks and Deep Learning (A) & Machine Learning Theory (A+) | Surveyed theory and practice of deep learning and Neural Networks; automatic differentiation, normalization, residual blocks, attention, model selection, validation. Replicated results of published study. |
| Advanced Topics in Computer Vision (A-) | Studied advanced computer vision topics, with focus on active vision and visual attention. |
| Simulation and Animation for Computer Games (A+) | Hermite and bezier curves, particle systems using verlet and/or other integration methods, Jacobian matrices and inverse kinematics. Implemented in C++ and OpenGL. |
| Research Methods for Human Computer Interaction (A) | Advanced concepts and technologies for human computer interaction research. Designed and conducted user study to evaluate user interfaces. |
| Data Analytics and Visualization (A) | Hands-on machine learning using WEKA, data analysis methods. |
Expertise
| Research and Domain Knowledge | Simulation, Computer Graphics, Machine Learning, Computer Vision, Visual Attention |
| Languages | Unity (C#), Python (PyTorch, Pandas, Numpy, SciPy), C++, R, OpenGL |
| Software | Unity, Blender, Unreal Engine 4, Autodesk Motion Builder |
| Platforms | Windows, Linux, Android, Github |
GET IN TOUCH
If you have any inquiries, please don't hesitate to contact me.