Revision 88d8aaf1c5885a75dce55fd1d648f43c3c3bece1 (click the page title to view the current version)

Lecture Notes - AIS2204 Maskinsyn

Chapter 1-2. Introduction and 3D Modelling (two weeks)

Dates 25-26 August + 1-2 September

# Session Notes Reading Keywords Status
1 Introduction Ma 2004:Ch 1 (Ch 2.1 and 2.3) Practical matters. Software installation. Recap of linear algebra. OK
2 3D Modelling Ma 2004:Ch 2, App A (SZ 2) 3D modelling, motion OK
3 3D Objects in Python Tutorials Homogeneous co-ordinates. General Rotations. 3D Transformations in Python OK
4 3D Modelling Part II Velocity transformations. Recap. Questions. To be adapted to class

Chapter 3. Image Formation (two weeks)

# Topic Reading Keywords Status
5 Image Formation Ma 2004:Ch 3-3.3.1 (SZ 6) projection, lens/camera OK
6 Camera Calibration Ma 2004:Ch 3.3-3.3.3 Calibration, Radial Distortion etc. OK
7 More Camera Mathematics Ma 2004:Ch 3.3-3.4 Radial Distortion, Tangential Distortion OK

Dates 8-9 and 15 September

Chapter 4. Feature Tracking (three weeks)

(Last year’s session numbers in parentheses.)

# Topic Reading Keywords Status
8 (new) Image Filters Convolution. Filters. Blurring. OK
9 (8-9) Corner Detection Ma 2004:Ch 4.3, 4.A (SZ 4) Calculate Gradient. Harris Feature Detector. OK
10 Tracking Features Ma 2004:Ch 4-4.2 Tracking of Features. Tracking of Edges. OK
11 (new) Project Tracker new Lecture Multiscale Detection OK
12 (16) SIFT Feature Matching. Feature Descriptor. OK
13-14 Self-Study Continue with Tracking Features - -
15 (11) Recap Status, review, repetition
16 (11) Edges Ma 2004:Ch 4.4 Canny, connected components, line fitting
  • 8 - 16 September
  • 9-12 - 22-23 and 29-30 September
  • 13-14 - 6-7 October - staff seminar - self-managed work only
  • 15-16 - 13-14 October - midterm - regular teaching

Chapter 5. Projective Reconstruction (two weeks ?)

# Topic Reading Keywords
17 Relative Pose Ma 2004:Ch 5.1 Triangulation. Relative Pose. Essential Matrix.
18 Eight-point algorithm Ma 2004:Ch 5.2 Calculate Essential Matrix
19 Study Technique Ma 2004:Ch 5.1 Proof reading.
20 3D Reconstruction Ma 2004:Ch 5.1-2
21 Planar Scenes Ma 2004:Ch 5.3
22 Epipolar Geometry Ma 2004:Ch 5.1-3
  1. Study Technique should may be needed earlier. We introduced some fragments of this in 2022, but should possibly do more of it.
  2. Relative Pose is a little messy. It serves covers two things.
    • triangulation is poorly covered in the textbook and the notes, but the students need a recap from basic calculus
    • the essential matrix is preparation for the next session.
  3. Generally, the preliminary steps of the exploratory exercises should have been premade, to save time for the students. We have made some improvements in 2022, but we should do more.
  4. Algorithm implementation is difficult and require sample solutions. Some have been added in 2022, but may have to be incorporated earlier in the course.
  5. We need more examples with complete calculations

New. Machine Learning

# Topic Reading Keywords
23-24 Neural Networks Szeliski 2022 Chapter 5 Training. Testing
25 Statistics Evaluation, Standard Deviation
26
  • Principles of Artificial Neural Networks
    • Graph Representation: Linear Combination + Non-Linear Activiation
      • Interpretation of Outputs
      • Loss Function
      • Optimisation Problem
    • Tensor Representation
      • Back-Propagation
    • Evaluation: Statistical Estimation and Hypothesis Test
  • Image recognition in PyTorch
    • Tutorials

Last Week. Summary and Miscellanea

# Topic Reading Keywords
23/2021 Distorted Space Ma 2004:Ch 6.1-2

Other Material.

Old Material.

# Topic Reading Keywords
24/2021 Stratified Reconstruction Ma 2005:Ch 6.3-4
25/2021 Partial Scene Information Ma 2005:Ch 6.5
26/2021 Real World Reconstruction Ma 2004:Ch 11
27/2021 Continuous Motion Ma 2004:Ch 5.4
  1. Ma 2004 Chapter 10. Partial Scene Knowledge
    • This is referenced as a building block in Chapter 11.
  2. Ma 2004 Chapter 11.4.
  3. Ma 2004 Chapter 11.5. Keywords texture, visualisation