CS 637: Computer Vision II

School of Electrical Engineering and Computer Science
Oregon State University
Corvallis, OR 97331

Instructor: Fuxin Li (KEC 2077)
Class Time: MW 10-11:50AM, Location: KEC 1001

Office Hours:
Fuxin Li: Monday 3PM, KEC 2077.

[registrar entry]

We use Canvas for discussions and grades. For technical questions (e.g., HW), first check if the same question has been asked on Canvas; if not, ask it there (we monitor the discussions). For non-technical questions (e.g., grading), send email to the instructor himself.

Midterm time: 5/10 in-class

Please note that the first class (4/2 10AM) is a recorded one. Please don't come to class on Monday. Sorry for this inconvenience.

Schedule/Slides/HWs


In a Nutshell

An introduction to recent advances in visual recognition, including object detection, semantic segmentation, optical flow, stereo matching, and human activity recognition. The course covers common formulations of these problems, including energy minimization on graphical models, and deep learning approaches to low- and high-level recognition tasks.

Prerequisites: CS 519-006 Deep Learning, CS 556 Computer Vision or equivalent knowledge recommended.


Textbooks:

There is no textbook for this class, we will be mostly reading papers. For the graphical model part, it is good to consult:

Koller and Friedman. Probabilistic Graphical Models: Principles and Techniques. MIT Press. 2009

However it is not mandatory and the course should be self-contained.

Learning Objectives of the Course:

    At the completion of the course, students will be able to:
  1. Have solid understanding of common concepts in low- and high-level visual recognition, such as object boundaries, optical flow, stereo, and semantic segmentation
  2. Have in-depth understanding of common formulations of problems in visual recognition
  3. Formulate a graphical model from the underlying vision problem, and solve its inference and learning
  4. Formulate a deep neural network suitable for solving a given vision problem, and derive its learning algorithm
  5. Use available software tools for graphical models and deep learning in visual recognition
  6. Solve visual recognition problems in real-world images and video

Assessment

  • 1 Midterm (20%)
  • 2 Assignments (20%) -- one of them would be paper review
  • In-class student presentation (20%): Each student will need to choose a paper from the designated paper list and present in-class
  • Final Project (40%) -- in groups of no more than three, including initial proposal presentation (10%, 5% on merit and 5% on clarity), final oral presentation (5%) and final written report (20% on merit, 5% on clarity)
  • Assignments

    There will be one written assignment on graphical models, as well as one assignment as a paper review.

    Final Project

    A self-designed project on a specific computer vision problem.

    NO FINAL EXAM.

    Final Project Presentation Time: Friday June 16th, 9:30AM

    Important Note: Please note that all homework (except the final project) should be your own work. Any collaboration that requires written communication is forbidden. You should also not copy answers from books or internet resources.


    Collaborations:

    You are encouraged to study together and discuss general strategies for solving problems but not at the level of written solutions. Please read the department's academic dishonesty policy for more details. You should not use any web sources for answering the homework questions unless explicitly instructed to do so.

    Fuxin Li