Göm meny

TSBB17 Visual Object Recognition and Detection HT2018

This course is worth 6 ECTS credits, which corresponds to approximately 160h of work per student. The time is divided among the following activities:

The course is given during the HT period. Note that not all the time slots in the course schedule will be used. A few spare slots have been reserved in case a lecture has to be moved. The actually planned lectures are listed below on this page, and any changes to this plan will be announced during the lectures, and on this page.

People

We both have our offices in the B-building, corridor D, first floor between entrances 25 and 27.

Documents

  • Detailed information and documents related to the two projects will be provided through the course repository hosted in IDAs GitLab (accessible to anyone with a LiU login).
  • Course description in Studieinfo (the web interface to Bilda)

Literature

  • Goodfellow et al., Deep Learning, MIT Press (2016).
    The book is available as full pdf. See also Book webpage.
  • Richard Szeliski, Computer Vision: Algorithms and Applications, Springer Verlag (2011).
    The book is available as an on campus e-book via the LiU library. See also Book webpage.
Additional material for side reading will be made available in the course repository.

Registration

If you intend to take the course but are not registered, make sure to register ASAP, using the Student portal. You need to be registered on the course to receive course email, and to have results input to Ladok. If you take the course but are not registered to any program at the University, please contact the course examiner in order to make sure that you receive email about the course.

Examination

The course has a written examination in English. The projects are examined using written reports and oral presentations.

Lecture schedule 2018

The lecture slides will be made available in the course repository after each lecture. The repository also contains additional relevant literature for side reading. All lectures and seminars will take place in the room Systemet at ISY.
Date,Time,Room Activity Teacher
September 3: 08.15-10
Systemet
Lecture 1
Introduction
Per-Erik Forssén
September 4: 10.15-12
Systemet
Lecture 2
Feature Descriptors
Per-Erik Forssén
September 5: 13.15-15
Systemet
Lecture 3
Convolutional Neural Networks: Introduction and Theory
Michael Felsberg
September 6: 10.15-12
Systemet
Lecture 4
Image Classification with Convolutional Neural Networks
Michael Felsberg
September 10: 08.15-10
Systemet
Lecture 5
Compound Descriptors and Metrics
Per-Erik Forssén
September 12: 13.15-15
Systemet
Lecture 6
Project 1: Visual Object Recognition
Per-Erik Forssén
September 14: 08.15-10
Systemet
Lecture 7
Visual Object Detection
Per-Erik Forssén
September 17: 08.15-10
Systemet
Lecture 8
Visual Object Tracking: Introduction
Michael Felsberg
September 18: 10.15-12
Systemet
Lecture 9
Discriminative Correlation Filters for Visual Tracking
Michael Felsberg
September 26: 13.15-15
Systemet
Seminar 1
Presentation of project 1
Per-Erik Forssén
September 26: 15.15-17
Systemet
Lecture 10
Project 2: Visual Object Tracking
Per-Erik Forssén
October 16: 10.15-12
Systemet
Seminar 2
Presentation of project 2
Per-Erik Forssén

Projects

The projects are conducted in groups of 4 or 3 students (in order of preference).

  • Project 1: Visual Object Recognition
    Introductory lecture: September 12: 13.15-15, Systemet
    Report due: September 23
    Presentation seminar: September 26: 13.15-15, Systemet
  • Project 2: Visual Object Tracking
    Introductory lecture: September 26: 15.15-17, Systemet
    Report due: October 9
    Presentation seminar: October 16: 10.15-12, Systemet

General resources

We suggest the use of following software:

  • PyTorch Deep learning framework for Python.
  • MatConvNet Deep Learning framework for Matlab.
  • Caffe Deep Learning library.
  • Theano Deep Learning library.
  • David Lowe's SIFT implementation for Matlab. This includes binary executables that are compiled into mex-code that runs from Matlab. If you don't know about Matlab mex-files ask your guide.
  • VLFeat has a a useful code library, both for Matlab and C/C++. For example, there is an alternative implementation of SIFT here, and also an implementation of MSER here. Both are made by Andrea Vedaldi.
  • The Visual Geometry Group at Oxford University maintains code for affine invariant region detectors, produced in cooperation with other groups.
  • LIBSVM A Library for Support Vector Machines (Matlab, Python).

Project repositories

Project code should be developed under versioning control, with changes tracked according to LiU-ID of the participating group members.

  • Option A: Project groups get their repositories from GITLab at IDA Note: this is not GitHub, and GitHub should not be used.
  • Option B: On request each project group can get access to a Subversion repository, with individual login by each project member.

    Documentation for Subversion can be found here:

    If you are using a Windows based computer, we recommend that you use the Tortoise client for Subversion.


Senast uppdaterad: 2018-09-05