Module Details

The information contained in this module specification was correct at the time of publication but may be subject to change, either during the session because of unforeseen circumstances, or following review of the module at the end of the session. Queries about the module should be directed to the member of staff with responsibility for the module.
Title Computer Vision
Code COMP338
Coordinator Dr T Do
Computer Science
Thanh-Toan.Do@liverpool.ac.uk
Year CATS Level Semester CATS Value
Session 2019-20 Level Three First Semester 15

Aims

To provide an introduction to the topic of Computer Vision.
To present fundamental problems in both 2D and 3D vision, and to explain a variety of classical and emerging approaches to overcome them.
To develop the practical skills necessary to build computer vision applications.


Learning Outcomes

(LO1) Demonstrate an understanding of the theoretical and practical aspects of image representations.

(LO2) Describe state-of-the-art techniques for image classification, image search, image segmentation, object detection, and object tracking.

(LO3) Describe the foundation of image formation with the pinhole camera model and how they project the 3D world to 2D images.

(LO4) Apply the principles of deep neural networks to various vision problems such as classification, detection, and semantic segmentation.

(LO5) Demonstrate and apply the practical skills necessary to build computer vision applications.


Syllabus

 

2D vision:
1. Course introduction, Computer vision overview
2. Linear Algebra for Computer Vision
3. Pixels and image representation
4. Image filters and edge detection
5. Local features and fitting (Local features, Harris corner detection, Scale invariant feature
transform, Image stitching and RANSAC)
6. Local features and fitting (Local features, Harris corner detection, Scale invariant feature
transform, Image stitching and RANSAC) (cont’d)
7. Python for Computer Vision
8. Segmentation: tree-based segmentation, spectral clustering, other superpixel methods.
9. Segmentation: tree-based segmentation, spectral clustering, other superpixel methods (cont’d)
10. Image classification overview and Bag of Features
11. Nearest neighbor and logistic regression (for image search and image classification)
12. Image classification and image search using advanced feature coding (First and second order
local feature aggreg ation, sparse coding).
13. Image classification and image search using advanced feature coding (First and second order
local feature aggregation, sparse coding) (cont’d)
14. Object detection (e.g., face and pedestrian) with sliding window approach, object detection
using part-based models.
15. Object detection (e.g., face and pedestrian) with sliding window approach, object detection
using part-based models. (cont’d)
16. Object detection (e.g., face and pedestrian) with sliding window approach, object detection
using part-based models. (cont’d)
17. Motion and tracking
18. Motion and tracking (cont’d)

3D vision
19. Image formulation, camera calibration
20. Image formulation, camera calibration (cont’d)
21. Multi-view geometry
22. Multi-view geometry (cont’d)
23. Multi-view geometry (cont’d)

Computer Vision in the era of deep learning
24. Neural Networks (perceptro n, multi-layer perceptron, activation functions, loss functions,
gradient descent, back-propagation).
25. Neural Networks (perceptron, multi-layer perceptron, activation functions, loss functions,
gradient descent, back-propagation). (cont’d)
26. Deep learning (principles, convolutional neural network, auto-encoder neural network,
recurrent neural network) and state-of-the-art deep network architectures for image
classification.
27. Deep learning (principles, convolutional neural network, auto-encoder neural network,
recurrent neural network) and state-of-the-art deep network architectures for image
classification. (cont’d)
28. Deep learning (principles, convolutional neural network, auto-encoder neural network,
recurrent neural network) and state-of-the-art deep network architectures for image
classification. (cont’d)
29. Deep learning for object detection and semantic segmentation
30. Deep learning for objec t detection and semantic segmentation (cont’d)


Teaching and Learning Strategies

Teaching Method 1 - Lecture
Description:
Attendance Recorded: Not yet decided
Notes: 3 lectures per week for 10 weeks

Teaching Method 2 - Laboratory Work
Description:
Attendance Recorded: Not yet decided


Teaching Schedule

  Lectures Seminars Tutorials Lab Practicals Fieldwork Placement Other TOTAL
Study Hours 30

    10

    40
Timetable (if known)              
Private Study 110
TOTAL HOURS 150

Assessment

EXAM Duration Timing
(Semester)
% of
final
mark
Resit/resubmission
opportunity
Penalty for late
submission
Notes
Final exam  2.5 hours    80       
CONTINUOUS Duration Timing
(Semester)
% of
final
mark
Resit/resubmission
opportunity
Penalty for late
submission
Notes
Programming Assignment on Image Alignment  10 hours over the co    10       
Programming Assignment on Deep Neural Networks  10 hours over the co    10       

Recommended Texts

Reading lists are managed at readinglists.liverpool.ac.uk. Click here to access the reading lists for this module.