skip to primary navigationskip to content

Department of Computer Science and Technology

Part II CST

 

Course pages 2022–23

Machine Visual Perception

Principal lecturer: Dr Cengiz Oztireli
Additional lecturer: Dr Christopher Town
Taken by: Part II CST
Code: MVP
Term: Michaelmas
Hours: 12 (Six 2-hour lectures)
Format: In-person lectures
Class limit: max. 30 students
Moodle, timetable

Aims

This course aims at introducing the theoretical foundations and practical techniques for machine perception, the capability of computers to interpret data resulting from sensor measurements. The course will teach the fundamentals and modern techniques of machine perception, i.e. reconstructing the real world starting from sensor measurements with a focus on machine perception for visual data. The topics covered will be image/geometry representations for machine perception, semantic segmentation, object detection and recognition, geometry capture, appearance modeling and acquisition, motion detection and estimation, human-in-the-loop machine perception, select topics in applied machine perception.

Machine perception/computer vision is a rapidly expanding area of research with real-world applications. An understanding of machine perception is also important for robotics, interactive graphics (especially AR/VR), applied machine learning, and several other fields and industries. This course will provide a fundamental understanding of and practical experience with the relevant techniques.

Learning outcomes

  • Students will understand the theoretical underpinnings of the modern machine perception techniques for reconstructing models of reality starting from an incomplete and imperfect view of reality.
  • Students will be able to apply machine perception theory to solve practical problems, e.g. classification of images, geometry capture.
  • Students will gain an understanding of which machine perception techniques are appropriate for different tasks and scenarios.
  • Students will have hands-on experience with some of these techniques via developing a functional machine perception system in their projects.
  • Students will have practical experience with the current prominent machine perception frameworks.

Syllabus

  • The fundamentals of machine learning for machine perception
  • Deep neural networks and frameworks for machine perception
  • Semantic segmentation of objects and humans
  • Object detection and recognition
  • Motion estimation, tracking and recognition
  • 3D geometry capture
  • Appearance modeling and acquisition
  • Select topics in applied machine perception

Assessment

  • A practical exercise, worth 20% of the mark. This will cover the basics and theory of machine perception and some of the practical techniques the students will likely use in their projects. This is individual work. No GPU hours will be needed for the practical work.
  • A machine perception project worth 80% of the marks:
    • Course projects will be selected by the students following the possible project themes proposed by the lecturer, and will be checked by the lecturer for appropriateness. 
    • The students will form groups of 2-3 to design, implement, report, and present a project to tackle a given task in machine perception.
    • The final mark will be composed of an implementation/report mark (60%) and a presentation mark (20%). Each team member will be evaluated based on her/his contribution.
    • Each project will have extensions to be completed only by the ACS students. Each student will write a different part of the report, whose author will be clearly marked. Each student will further summarise her/his contributions to the project in the same report.

Recommended Reading List

  • Computer Vision: Algorithms and Applications, Richard Szeliski, Springer, 2010.
  • Deep Learning, Ian Goodfellow, Yoshua Bengio, and Aaron Courville, MIT Press, 2016.
  • Machine Learning and Visual Perception, Baochang Zhang, De Gruyter, 2020.