Announcements
- Please see the Quercus page for this course for complete details.
Overview
A capstone course is for showcasing your existing skills, and much of the course involves self-learning or learning with your team under the guidance and supervision of instructors. You are in charge of your own learning and progress in this course, and the instructors will provide a push in the right direction as needed (and not to provide traditional lectures/assignments/etc. as you would find in other courses).
This year the course will focus on Machine Learning for Vision. The use of machine learning, and especially deep learning, has led to great progress in solving many difficult problems in machine vision. Example problems include object detection and localization, image segmentation, texture synthesis, scene labelling, colorization, learning artistic style, visual question answering, motion estimation, computational photography, depth estimation, 3D reconstruction, and many others.This capstone design course strives for novel approaches and results in using machine learning to solve problems in machine vision, such as employing a new or advanced combination of a dataset and a machine learning method. Students will work in teams of size 2-4. Teams will be formed in class during the first two to three weeks of the term. Very excellent teams might come up with publishable methods and results (which are an impressive addition to any resume, especially if you are applying to graduate school).
Prerequisites:
This course assumes that students have significant programming experience and a solid foundation in the theory and practice of machine learning, including neural networks. Basic methods in machine learning include linear regression and classification, logistic regression, K nearest neighbours, decision trees, probabilistic models, ensemble methods, dimensionality reduction and clustering. In addition, any project in machine learning requires a mastery of training, validation and testing, overfitting and under-fitting, and supervised and unsupervised learning. Most work in machine vision also requires convolutional neural networks (CNNs).
The formal prerequisites are CSC209 and csc311. (Ask for permission from the instructor if any prerequisites are missing.) No special knowledge of machine vision is required.