Computer Science
Faculty of Science
Ontario Tech University (formerly UOIT)
SCI4000, 2000 Simcoe St. N., Oshawa, ON L1G OC5 Canada
faisal.qureshi@ontariotechu.ca
(905) 721-8668 x 3626
Messages
Please read this if you are considering doing an honours thesis with me.
Please read this if you are considering doing graduate studies under my supervision.
News
About
I am a professor of computer science at Ontario Tech University, where I lead the Visual Computing Lab. My lab develops computational models of visual perception for autonomous systems that must act purposefully in a world that is difficult to model. Our current work spans computer vision, machine learning and hyperspectral imaging.
Current research problems in the lab include:
- hyperspectral image analysis, unmixing and compression;
- efficient vision transformers;
- visual anomaly detection;
- human pose and 3D mesh recovery; and
- visual sensor networks and virtual vision (3D software laboratories for camera networks research).
From 2020 to 2025 I held a guest professorship at Mid Sweden University. I hold adjunct appointments at the Information Technology University (Lahore) and the Institute of Space Technologies (Islamabad), and a status-only appointment at the University of Toronto. I am Vice President of the Canadian Image Processing and Pattern Recognition Society, which gave me its Lifetime Achievement (Service) Award in 2025. In 2026 I was elected a full member of Sigma Xi.
more >>>
Research
hyperspectral imagery
ATLAS: Hyperspectral Image Compression using Gaussian Splatting
A study of compressing hyperspectral images using Gaussian Splatting. What works? What doesn't?
Hyperspectral Pixel Unmixing with Bundled Endmembers
A model for hyperspectral pixel unmixing that explicitly accounts for variability in endmembers spectra.
Latent Dirichlet VAE for Hyperspectral Pixel Unmixing
Latent Dirichlet VAE proposes a new view of hyperspectral pixel unmixing where the VAE's latent representation follows a Dirichlet distribution and models per-pixel abundances.
Implicit Neural Representations for HSI Compression
Using implicit neural representations for hyperspectral image compression.
Hyperspectral Pixel Classification
A systematic study of the effects of spectral compression on hyperspectral pixel classification.
Hyperspectral Image Segmentation
Optimal scale selection for object-based image analysis and interpretation in hyperspectral images.
Iterative LDVAE for Hyperspectral Pixel Unmixing
A method for hyperspectral pixel unmixing that does not require labelled per-pixel "training" data.
computer vision
CascadedViT
A lightweight and compute-efficient vision transformer architecture featuring a novel feedforward network design called Cascaded-Chunk Feed Forward Network. As compared to EfficientViT variants, across various model sizes, CascadedViT family consistently exhibits the lowest energy consumption, making it suitable for deployment on battery-constrained devices such as mobile phones and drones.
A Modular Approach to Image Matting
A method that bridges the gap between the performance of trimap-based networks and the convenience of trimap-free networks.
Joint Spatial and Layer Attention for Convolutional Networks
What and where attention for convolutional neural networks
Tracking Worms
Microfilaria motion analysis to study the effects of environmental factors on worm movement.
Cost-Volume Filtering for Optical Flow Estimation
Cost-volume filtering for fast optical flow estimation
Cost-Volume Filtering for Stereo Depth Estimation
Accelerated cost-volume filtering for depth analysis
Computer Vision Pipelines
A formal algebraic framework for the mathematical description of computer vision pipelines.
Tracking Blood Droplets in 3D
Extracting three-dimensional flight trajectories of liquid droplets from video data.
Road Boundary Detection
Road boundary detection adverse lighting and environmental conditions.
A Residual-Dyad Encoder Discriminator Network for Remote Sensing Image Matching
A new deep learning unit, residual-dyad, for remote sensing image matching.
Satellite Rendezvous and Docking
Cognitive/behavior visual serving controllers for capturing free-flying uncooperative satellites.
Video Driven Crowd Synthesis
Motion vectors extracted from crowd videos are used to synthesize artificial crowds.
Feet Localization for Autonomous Wheelchair Navigation
Caregiver feet localization from wheelchair mounted cameras for autonomous navigation.
Bird Detection in Wind Farms
A temporal boosted YOLOv4 model for bird detection around wind farms.
anomaly detection
Visual Anomaly Detection Package
A web-based system for deploying anomalib for visual anomaly detection.
Multivariate Anomaly Detection
On aggregated vs. separating scoring methods for multivariate anomaly detection tasks.
camera networks
Virtual Vision
Advanced computer graphics and vision simulation technologies to serve the needs of camera networks research.
Multi-Tasking PTZ Cameras
Behavior-based controllers for passive and active PTZ cameras.
ConvNeXT for Person Re-ID
Using ConvNeXT as feature extractor with IICS/IIDS framework for person re-identification
other
Neural Networks Solving PDEs
Neural networks learn general representations when trained to solve differential equations.
Datasets
UofT HSI 301
University of Toronto Hyperspectral-301 dataset consisting of three 301-channel hyperspectral images covering (1) suburban, (2) urban, and (3) forest landtypes.
OnTech HSISyn 21
Hyperspectral pixel unmixing benchmark. A synthetic hyperspectral image with nine endmembers and per-pixel abundances.
Cuprite Synthetic
Synthetic hyperspectral dataset comprise materials (endmembers) found in Cuprite dataset. Unmixing models trained on this dataset can be used to perform unmixing on Cuprite dataset via transfer learning.
Teaching
Notes
Computer Vision
Covers topics discussed in both undergraduate and graduate computer vision courses.
Programming in C++
Covers topics discussed in a first-year C++ programming courses.
CNNs for Object Detection
Discusses the unreasonable effectiveness of convolutional neural networks for object detection.
Deep Learning Primer
Discusses topics, such as t-SNE, gradient descent, and variational and conditional autoencoders.
Simulation and Modeling
Discusses topics covered in the undergraduate simulation and modeling course.
Courses
Current
CSCI 3240U
[Web]
Computer Vision I
Fall 2026
CSCI 5520G
[Web]
Computer Vision and Deep Learning (graduate course)
Fall 2026
Past
CSCI 1061U
[Web]
Programming Workshop II
Last offered in Winter 2021
CSCI 2050U
[Web]
Computer Architecture I
Last offered in Winter 2026
CSCI 3010U
[Web]
Simulation and Modeling
Last offered in Winter 2026
CSCI 3240U
[Web]
Computational Photography
Last offered in Fall 2025 (Changed to Computer Vision I)
CSCI 4220U
[Web]
Computer Vision
Last offered in Winter 2026
CSCI 4440U
[Web]
Topics in Computer Science 1 (Human Action Recognition)
Last offered in Fall 2024
CSCI 5550G
[Web]
Topics in Digital Media
Last offered in Fall 2018
CSCI 5770G
[Web]
Machine Learning (graduate course)
Last offered in Winter 2023
MCSC 6230G/7230G
[Web]
Machine Learning and Deep Learning (Adv. Topics in HPC)
Last offered in Fall 2017
Coding
From time to time I find myself compelled to do some coding. Below I list some open source software that I have written that you may find useful.
- Webify, a python wrapper around pandoc to generate websites and pdf documents from markdown, yaml and mustache files.
[Web] [Github] - Distributed virtual vision simulator
[Web] [Github] - A python utility to export itunes playlists
[Github] - A python utility for setting captions of images based upon their filenames
[Github]
Nostalgia
Infomate DOS (c. 95/96)
A DOS based application for forward and inverse kinematics for a robotic manipulator.
A Raytracer in Java (c. 99)
My Java-based ray tracer that I wrote for a course assignment back in 1999.
Publications
Last updated 2026-11-03 08:52
© Webify 4.1