fruittree/csc496_2019f

Deep learning in computer vision

CSC496_2019F Deep learning in computer vision

Instructor: Bei Xiao, American University Location: DMTI Room 110 Time: Monday, Thursday 2:30pm-3:45pm Office: DMTI Room 204 First Class: August 26th

Office hour: check syllabus

General Information

Course Objective

This course provides an overview of “deep neural network” (AKA deep learning) and related methods with applications in computer vision. First, we will overview (continue after CSC 476) some fundamental topics in computer vision such as image formation, camera models, color, multiscale pyramids, and statistical modeling of images. Second, we will introduce basic machine learning methods and neural network architectures. Finally, we dive into a variety of deep learning-based applications in computer vision. Students learn to implement, train, and debug off-the shelf and their own neural networks and gain a detailed understanding of the cutting-edge research in computer vision. Topics include visual data representation, mid-level vision, image parsing, image classification and synthesis with architectures such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), generative-advisory networks (GANs), and Variational Autoencoders (VAEs).

Perquisites

1. Proficiency in Python. If you have plenty of coding experieces with MATLAB, C++, is fine too.
2. College Calculus
3. Basic probability, statistics andlinear algebra.

Textbooks and references

1. (Required) R. Szeliski, Computer Vision: Algorithms and Applications
, available at (http://szeliski.org/Book/)

2. (Required) I. Goodfellow, Deep learning (http://www.deeplearningbook.org)

Other references (especially useful if you haven’t taken CSC476):

3. Linear Algebra 3.1 Stanford CS229 review (http://cs229.stanford.edu/section/cs229-linalg.pdf) 3.2 Immersive Math, linear algebra (http://immersivemath.com/ila/index.html)

4. Python/Numpy tutorials (https://sites.engineering.ucsb.edu/~shell/che210d/numpy.pdf)

5. Probability and statistics, seeing theory (https://seeing-theory.brown.edu/)

6. Tensorflow and machine learning (https://www.tensorflow.org/tutorials/keras)

Course Policy

Attendence policy

Missing one class without written request will result in 2% reduction in attendance score. However, students can be absent for class and arrange for make-up exams if they have written proof of religious holidays and documented disabilities. Athlete who would miss class due to sports events must send written form to instructor at least 24 hours before the class.

Homework late policy

Projects are due 11:59pm of the due date. You usually have two weeks to finish the assigned project.

We do not accept late submissions for this course. Homework is typically due 11:59pm. The submission deadline has a 24-hour soft cut-off; after midnight, submissions are penalized 5% per hour late round up to the next hour (automated by Blackboard). So, you turned it in 2:59am. You will get 3 hours penalty with a 15% point reduction. There is no negotiation about this.

Exam Policy

Mid-term exam will be announced at least one week ahead of time. If you have special needs, you need to notify me at least 7 days before to arrange the test be performed off-class in the exam center. Missed exams and quizzes cannot be made up.

Email Policy

You can email me if you have questions regarding clarification of lectures and homeworks. But you must write to me at least 48 hours to expect an answer. No homework is accepted via Email. Everything must be uploaded onto Blackboard. Elaborate homework-related questions are restricted to office hours.

Grading

65% homework projects, 10% mid-term exam, 15% Final project, 5% in-class quiz, 5% attendances. We sometimes offer extra credits for additional features in homework and projects.

Course Schedule

Date	Lecture&Topic	Course Materials	Assignments
Mon, August 26	Lecture 1: Course Introduction	Szeliski 1, Goodfellow 2:Linear Algebra review	Exercise 1: set up virtual box, Numpy primer, jupyter notebook tutorial
Thursday, August 29	Lecture 2: A simple visual system	Blockworld: a simple vision system	VirtualBox Tutorial
Thursday, September 5	Lecture 3: pinhoel camera	Cameras	Homework 1:simple vision system
Monday, September 9	Lecture 4: Lens; Filtering	SignalProcessing
Thursday, September 13	Lecture 5: Linear Filtering, 2D convolution		Homework 2: Numpy/Scipy Exercises
Monday, September 16	Lecture 6: Signal Processing 2: Fourier transform	Szeliski Chapter 3.2,3.3
Monday, September 23	Lecture 7: Convolution in Fourier Domain		Homework 3: Pin-hole camera
Monday, September 26	Lecture 8: Temporal Filters	TemporalProcessing