EE462, EE9SO25, EE9CS728:
Machine Learning for Computer Vision 

 

    Course Aims

• The course aims to introduce the concepts, theories and state-of-the-art algorithms for visual learning and recognition. The first half of the module, for formulations and theories of machine learning techniques, consists of clustering, discriminative classifier learning, and probabilistic generative models. The latter half leads to the topics of visual recognition by the machine learning techniques learnt, including object detection/categorisation, segmentation, face and pose recognition. Each topic is illustrated with case studies.

• Syllabus:

Machine learning techniques: Prob. theory, Polynimial curve fitting, Clustering and EM, Boosting, Randomised decision forests, Maximum margin classifier, Markov random fields, Manifold learning and Gaussian process.

Visual learning and recognition: Image clustering, Object detection, Bag of words, Object recognition/categorisation, Segmentation, Face recognition and Pose estimation.


    Course Formats

• Autumn term 2014: every Monday, at 9-11am (2hours), in the room 509A, EEE, from 13th Oct to 8th Dec

• Assessment: 100% coursework during the term time

• Prerequisite: None

• Lecturer:

Dr Tae-Kyun Kim
office hours: Mon 11am-12noon, 1017
e-mail: tk.kim@imperial.ac.uk

• GTAs:

Dr Rigas Kouskouridas
office hours: Tue 11am-12noon, 1003
e-mail: r.kouskouridas@imperial.ac.uk

Guillermo Garcia
office hours: Wed 3-4pm, 1008d
e-mail: g.garcia-hernando@imperial.ac.uk

Mang Shao
office hours: Thu 3-4pm, 1008d
e-mail: mang.shao08@imperial.ac.uk

• Course website: http://www.iis.ee.ic.ac.uk/~tkkim/mlcv.htm


    Course Schedule and Lecture Notes
    (lectures notes to be updated)

• Lecture 1-2: Course Overview / Background

   [Lecture note in ppt and pdf]


• Lecture 3-4: Clustering and EM

   [Lecture note in ppt and pdf]


• Lecture 5-6: Object Detection / Boosting

   [Lecture note in ppt and pdf]


• Lecture 7-8: Random Forest and Applications

   [Lecture note in ppt and pdf]


• Lecture 9-10: Object Recognition, Categorisation / Maximum Margin Classifier

   [Lecture note in ppt]


• Lecture 11-12: Segmentation / Markov Random Fields

   [Lecture note in ppt]


• Lecture 13-14: Face Recognition / Manifold Learning

   [Lecture note in ppt]


• Lecture 15-16: Pose Estimation / Gaussian Process

   [Lecture note in ppt]


• Lecture 17: Summary


Resources:
   [Statistical Pattern Recognition Toolbox for Matlab]
   [Matlab demo code for image denoising by MRFs]
   [Matlab toolbox for GP]


    Coursework Schedule and Material

See the coursework guidelines. The coursework material are released on time, not earlier, at this website and Blackboard.

All course works require Matlab programming. Some questions marked* are about theories, answering them does not require Matlab programming.

Courseworks are due about every two weeks. Refer to the schedules and mark percentages below to help your planning. They are tentative, and subject to minor changes depending on our progress. The final score will not be the sum of all marks above. They will be moderated, as appropriate.


• Coursework 1: 15%, release: 13 Oct 2014, due: 26 Oct 2014 (11:59pm)

    Contents: Matlab basics, Matlab image interfaces, Lecture 1-2
    [Coursework material]


• Coursework 2: 25%, release: 27 Oct 2014, due: 9 Nov 2014 (11:59pm)

    Contents: Lecture 3-4,5-6
    [Coursework instruction and data/code]


• Coursework 3: 25%, release: 10 Nov 2014, due: 30 Nov 2014 (11:59pm)

    Contents: Lecture 7, 9-11
    [Coursework material]


• Coursework 4: 25%, release: 01 Dec 2014, due: 14 Dec 2014 (11:59pm)

    Contents: Lecture 12, 13-14, 15-16
    [Coursework material]


• Individual interview: 10%, 15 – 19 Dec 2014 (at office hours)

    Contents: All courseworks and lectures


The courseworks (reports and Matlab codes) should be submitted to Blackboard electronically. And drop a hardcopy of the report in the homework dropbox at EEE 1017.
Write your full name and CID number on the top of the first page.

Penalty on late submissions is applied, by 10% mark deduction for each day.



    *Warning on plagiarism! You will receive 0 mark when you are found to cheat, to copy others’ reports and codes, rather than using provided codes or Matlab default toolboxes. If you are uncertain, please contact the GTAs.

    *Please register the course, announcements will be made by the course email-list.